diff --git a/cookbook/litellm_proxy_server/cli_token_usage.py b/cookbook/litellm_proxy_server/cli_token_usage.py index e6b3744019c..c9c91e3283b 100644 --- a/cookbook/litellm_proxy_server/cli_token_usage.py +++ b/cookbook/litellm_proxy_server/cli_token_usage.py @@ -3,7 +3,7 @@ Example: Using CLI token with LiteLLM SDK This example shows how to use the CLI authentication token -in your Python scripts after running `litellm-proxy login`. +in your Python scripts after running `lite login`. """ from textwrap import indent @@ -22,7 +22,7 @@ def main(): api_key = litellm.get_litellm_gateway_api_key() if not api_key: - print("āŒ No CLI token found. Please run 'litellm-proxy login' first.") + print("āŒ No CLI token found. Please run 'lite login' first.") return print("āœ… Found CLI token.") @@ -58,6 +58,6 @@ if __name__ == "__main__": main() print("\nšŸ’” Tips:") - print("1. Run 'litellm-proxy login' to authenticate first") + print("1. Run 'lite login' to authenticate first") print("2. Replace 'https://your-proxy.com' with your actual proxy URL") print("3. The token is stored in your OS keychain, or in ~/.litellm/token.json when there is none") diff --git a/cookbook/litellm_proxy_server/grafana_dashboard/dashboard_1/grafana_dashboard.json b/cookbook/litellm_proxy_server/grafana_dashboard/dashboard_1/grafana_dashboard.json deleted file mode 100644 index 269c1ea5a43..00000000000 --- a/cookbook/litellm_proxy_server/grafana_dashboard/dashboard_1/grafana_dashboard.json +++ /dev/null @@ -1,614 +0,0 @@ -{ - "annotations": { - "list": [ - { - "builtIn": 1, - "datasource": { - "type": "grafana", - "uid": "-- Grafana --" - }, - "enable": true, - "hide": true, - "iconColor": "rgba(0, 211, 255, 1)", - "name": "Annotations & Alerts", - "target": { - "limit": 100, - "matchAny": false, - "tags": [], - "type": "dashboard" - }, - "type": "dashboard" - } - ] - }, - "description": "", - "editable": true, - "fiscalYearStartMonth": 0, - "graphTooltip": 0, - "id": 2039, - "links": [], - "liveNow": false, - "panels": [ - { - "datasource": { - "type": "prometheus", - "uid": "${DS_PROMETHEUS}" - }, - "fieldConfig": { - "defaults": { - "color": { - "mode": "palette-classic" - }, - "custom": { - "axisCenteredZero": false, - "axisColorMode": "text", - "axisLabel": "", - "axisPlacement": "auto", - "barAlignment": 0, - "drawStyle": "line", - "fillOpacity": 0, - "gradientMode": "none", - "hideFrom": { - "legend": false, - "tooltip": false, - "viz": false - }, - "lineInterpolation": "linear", - "lineWidth": 1, - "pointSize": 5, - "scaleDistribution": { - "type": "linear" - }, - "showPoints": "auto", - "spanNulls": false, - "stacking": { - "group": "A", - "mode": "none" - }, - "thresholdsStyle": { - "mode": "off" - } - }, - "mappings": [], - "thresholds": { - "mode": "absolute", - "steps": [ - { - "color": "green", - "value": null - }, - { - "color": "red", - "value": 80 - } - ] - }, - "unit": "s" - }, - "overrides": [] - }, - "gridPos": { - "h": 8, - "w": 12, - "x": 0, - "y": 0 - }, - "id": 10, - "options": { - "legend": { - "calcs": [], - "displayMode": "list", - "placement": "bottom", - "showLegend": true - }, - "tooltip": { - "mode": "single", - "sort": "none" - } - }, - "targets": [ - { - "datasource": { - "type": "prometheus", - "uid": "${DS_PROMETHEUS}" - }, - "editorMode": "code", - "expr": "histogram_quantile(0.99, sum(rate(litellm_self_latency_bucket{self=\"self\"}[1m])) by (le))", - "legendFormat": "Time to first token", - "range": true, - "refId": "A" - } - ], - "title": "Time to first token (latency)", - "type": "timeseries" - }, - { - "datasource": { - "type": "prometheus", - "uid": "${DS_PROMETHEUS}" - }, - "fieldConfig": { - "defaults": { - "color": { - "mode": "palette-classic" - }, - "custom": { - "axisCenteredZero": false, - "axisColorMode": "text", - "axisLabel": "", - "axisPlacement": "auto", - "barAlignment": 0, - "drawStyle": "line", - "fillOpacity": 0, - "gradientMode": "none", - "hideFrom": { - "legend": false, - "tooltip": false, - "viz": false - }, - "lineInterpolation": "linear", - "lineWidth": 1, - "pointSize": 5, - "scaleDistribution": { - "type": "linear" - }, - "showPoints": "auto", - "spanNulls": false, - "stacking": { - "group": "A", - "mode": "none" - }, - "thresholdsStyle": { - "mode": "off" - } - }, - "mappings": [], - "thresholds": { - "mode": "absolute", - "steps": [ - { - "color": "green", - "value": null - }, - { - "color": "red", - "value": 80 - } - ] - }, - "unit": "currencyUSD" - }, - "overrides": [ - { - "matcher": { - "id": "byName", - "options": "7e4b0627fd32efdd2313c846325575808aadcf2839f0fde90723aab9ab73c78f" - }, - "properties": [ - { - "id": "displayName", - "value": "Translata" - } - ] - } - ] - }, - "gridPos": { - "h": 8, - "w": 12, - "x": 0, - "y": 8 - }, - "id": 11, - "options": { - "legend": { - "calcs": [], - "displayMode": "list", - "placement": "bottom", - "showLegend": true - }, - "tooltip": { - "mode": "single", - "sort": "none" - } - }, - "targets": [ - { - "datasource": { - "type": "prometheus", - "uid": "${DS_PROMETHEUS}" - }, - "editorMode": "code", - "expr": "sum(increase(litellm_spend_metric_total[30d])) by (hashed_api_key)", - "legendFormat": "{{team}}", - "range": true, - "refId": "A" - } - ], - "title": "Spend by team", - "transformations": [], - "type": "timeseries" - }, - { - "datasource": { - "type": "prometheus", - "uid": "${DS_PROMETHEUS}" - }, - "fieldConfig": { - "defaults": { - "color": { - "mode": "palette-classic" - }, - "custom": { - "axisCenteredZero": false, - "axisColorMode": "text", - "axisLabel": "", - "axisPlacement": "auto", - "barAlignment": 0, - "drawStyle": "line", - "fillOpacity": 0, - "gradientMode": "none", - "hideFrom": { - "legend": false, - "tooltip": false, - "viz": false - }, - "lineInterpolation": "linear", - "lineWidth": 1, - "pointSize": 5, - "scaleDistribution": { - "type": "linear" - }, - "showPoints": "auto", - "spanNulls": false, - "stacking": { - "group": "A", - "mode": "none" - }, - "thresholdsStyle": { - "mode": "off" - } - }, - "mappings": [], - "thresholds": { - "mode": "absolute", - "steps": [ - { - "color": "green", - "value": null - }, - { - "color": "red", - "value": 80 - } - ] - } - }, - "overrides": [] - }, - "gridPos": { - "h": 9, - "w": 12, - "x": 0, - "y": 16 - }, - "id": 2, - "options": { - "legend": { - "calcs": [], - "displayMode": "list", - "placement": "bottom", - "showLegend": true - }, - "tooltip": { - "mode": "single", - "sort": "none" - } - }, - "targets": [ - { - "datasource": { - "type": "prometheus", - "uid": "${DS_PROMETHEUS}" - }, - "editorMode": "code", - "expr": "sum by (model) (increase(litellm_requests_metric_total[5m]))", - "legendFormat": "{{model}}", - "range": true, - "refId": "A" - } - ], - "title": "Requests by model", - "type": "timeseries" - }, - { - "datasource": { - "type": "prometheus", - "uid": "${DS_PROMETHEUS}" - }, - "fieldConfig": { - "defaults": { - "color": { - "mode": "thresholds" - }, - "mappings": [], - "noValue": "0", - "thresholds": { - "mode": "absolute", - "steps": [ - { - "color": "green", - "value": null - }, - { - "color": "red", - "value": 80 - } - ] - } - }, - "overrides": [] - }, - "gridPos": { - "h": 7, - "w": 3, - "x": 0, - "y": 25 - }, - "id": 8, - "options": { - "colorMode": "value", - "graphMode": "area", - "justifyMode": "auto", - "orientation": "auto", - "reduceOptions": { - "calcs": [ - "lastNotNull" - ], - "fields": "", - "values": false - }, - "textMode": "auto" - }, - "pluginVersion": "9.4.17", - "targets": [ - { - "datasource": { - "type": "prometheus", - "uid": "${DS_PROMETHEUS}" - }, - "editorMode": "code", - "expr": "sum(increase(litellm_llm_api_failed_requests_metric_total[1h]))", - "legendFormat": "__auto", - "range": true, - "refId": "A" - } - ], - "title": "Faild Requests", - "type": "stat" - }, - { - "datasource": { - "type": "prometheus", - "uid": "${DS_PROMETHEUS}" - }, - "fieldConfig": { - "defaults": { - "color": { - "mode": "palette-classic" - }, - "custom": { - "axisCenteredZero": false, - "axisColorMode": "text", - "axisLabel": "", - "axisPlacement": "auto", - "barAlignment": 0, - "drawStyle": "line", - "fillOpacity": 0, - "gradientMode": "none", - "hideFrom": { - "legend": false, - "tooltip": false, - "viz": false - }, - "lineInterpolation": "linear", - "lineWidth": 1, - "pointSize": 5, - "scaleDistribution": { - "type": "linear" - }, - "showPoints": "auto", - "spanNulls": false, - "stacking": { - "group": "A", - "mode": "none" - }, - "thresholdsStyle": { - "mode": "off" - } - }, - "mappings": [], - "thresholds": { - "mode": "absolute", - "steps": [ - { - "color": "green", - "value": null - }, - { - "color": "red", - "value": 80 - } - ] - }, - "unit": "currencyUSD" - }, - "overrides": [] - }, - "gridPos": { - "h": 7, - "w": 3, - "x": 3, - "y": 25 - }, - "id": 6, - "options": { - "legend": { - "calcs": [], - "displayMode": "list", - "placement": "bottom", - "showLegend": true - }, - "tooltip": { - "mode": "single", - "sort": "none" - } - }, - "targets": [ - { - "datasource": { - "type": "prometheus", - "uid": "${DS_PROMETHEUS}" - }, - "editorMode": "code", - "expr": "sum(increase(litellm_spend_metric_total[30d])) by (model)", - "legendFormat": "{{model}}", - "range": true, - "refId": "A" - } - ], - "title": "Spend", - "type": "timeseries" - }, - { - "datasource": { - "type": "prometheus", - "uid": "${DS_PROMETHEUS}" - }, - "fieldConfig": { - "defaults": { - "color": { - "mode": "palette-classic" - }, - "custom": { - "axisCenteredZero": false, - "axisColorMode": "text", - "axisLabel": "", - "axisPlacement": "auto", - "barAlignment": 0, - "drawStyle": "line", - "fillOpacity": 0, - "gradientMode": "none", - "hideFrom": { - "legend": false, - "tooltip": false, - "viz": false - }, - "lineInterpolation": "linear", - "lineWidth": 1, - "pointSize": 5, - "scaleDistribution": { - "type": "linear" - }, - "showPoints": "auto", - "spanNulls": false, - "stacking": { - "group": "A", - "mode": "none" - }, - "thresholdsStyle": { - "mode": "off" - } - }, - "mappings": [], - "thresholds": { - "mode": "absolute", - "steps": [ - { - "color": "green", - "value": null - }, - { - "color": "red", - "value": 80 - } - ] - } - }, - "overrides": [] - }, - "gridPos": { - "h": 7, - "w": 6, - "x": 6, - "y": 25 - }, - "id": 4, - "options": { - "legend": { - "calcs": [], - "displayMode": "list", - "placement": "bottom", - "showLegend": true - }, - "tooltip": { - "mode": "single", - "sort": "none" - } - }, - "targets": [ - { - "datasource": { - "type": "prometheus", - "uid": "${DS_PROMETHEUS}" - }, - "editorMode": "code", - "expr": "sum(increase(litellm_total_tokens_total[5m])) by (model)", - "legendFormat": "__auto", - "range": true, - "refId": "A" - } - ], - "title": "Tokens", - "type": "timeseries" - } - ], - "refresh": "1m", - "revision": 1, - "schemaVersion": 38, - "style": "dark", - "tags": [], - "templating": { - "list": [ - { - "current": { - "selected": false, - "text": "prometheus", - "value": "edx8memhpd9tsa" - }, - "hide": 0, - "includeAll": false, - "label": "datasource", - "multi": false, - "name": "DS_PROMETHEUS", - "options": [], - "query": "prometheus", - "queryValue": "", - "refresh": 1, - "regex": "", - "skipUrlSync": false, - "type": "datasource" - } - ] - }, - "time": { - "from": "now-1h", - "to": "now" - }, - "timepicker": {}, - "timezone": "", - "title": "LLM Proxy", - "uid": "rgRrHxESz", - "version": 15, - "weekStart": "" - } \ No newline at end of file diff --git a/cookbook/litellm_proxy_server/grafana_dashboard/dashboard_1/readme.md b/cookbook/litellm_proxy_server/grafana_dashboard/dashboard_1/readme.md deleted file mode 100644 index 1f193aba702..00000000000 --- a/cookbook/litellm_proxy_server/grafana_dashboard/dashboard_1/readme.md +++ /dev/null @@ -1,6 +0,0 @@ -## This folder contains the `json` for creating the following Grafana Dashboard - -### Pre-Requisites -- Setup LiteLLM Proxy Prometheus Metrics https://docs.litellm.ai/docs/proxy/prometheus - -![1716623265684](https://github.com/BerriAI/litellm/assets/29436595/0e12c57e-4a2d-4850-bd4f-e4294f87a814) diff --git a/cookbook/litellm_proxy_server/grafana_dashboard/dashboard_all_metrics/grafana_dashboard.json b/cookbook/litellm_proxy_server/grafana_dashboard/dashboard_all_metrics/grafana_dashboard.json new file mode 100644 index 00000000000..d8cb122417a --- /dev/null +++ b/cookbook/litellm_proxy_server/grafana_dashboard/dashboard_all_metrics/grafana_dashboard.json @@ -0,0 +1,6312 @@ +{ + "annotations": { + "list": [ + { + "builtIn": 1, + "datasource": { + "type": "grafana", + "uid": "-- Grafana --" + }, + "enable": true, + "hide": true, + "iconColor": "rgba(0, 211, 255, 1)", + "name": "Annotations & Alerts", + "type": "dashboard" + } + ] + }, + "description": "Every litellm_* Prometheus metric the LiteLLM proxy emits, one panel per metric family, grouped by theme.", + "editable": true, + "fiscalYearStartMonth": 0, + "graphTooltip": 1, + "id": null, + "links": [], + "panels": [ + { + "collapsed": false, + "gridPos": { + "h": 1, + "w": 24, + "x": 0, + "y": 0 + }, + "id": 1, + "panels": [], + "title": "Proxy traffic", + "type": "row" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Total number of requests made to the proxy server - track number of client side requests", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "short" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 0, + "y": 1 + }, + "id": 2, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_proxy_total_requests_metric_total[$__rate_interval])) by (status_code)", + "legendFormat": "{{status_code}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_proxy_total_requests_metric rate", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Total number of failed responses from proxy - the client did not get a success response from litellm proxy", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "short" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 12, + "y": 1 + }, + "id": 3, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_proxy_failed_requests_metric_total[$__rate_interval])) by (exception_class)", + "legendFormat": "{{exception_class}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_proxy_failed_requests_metric rate", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "deprecated - use litellm_proxy_total_requests_metric. Total number of LLM calls to litellm - track total per API Key, team, user", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "short" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 0, + "y": 9 + }, + "id": 4, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_requests_metric_total[$__rate_interval])) by (model)", + "legendFormat": "{{model}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_requests_metric rate", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "deprecated - use litellm_proxy_failed_requests_metric", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "short" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 12, + "y": 9 + }, + "id": 5, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_llm_api_failed_requests_metric_total[$__rate_interval])) by (model)", + "legendFormat": "{{model}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_llm_api_failed_requests_metric rate", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Number of HTTP requests currently in-flight on this uvicorn worker", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "short" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 0, + "y": 17 + }, + "id": 6, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "litellm_in_flight_requests", + "legendFormat": "__auto", + "range": true, + "refId": "A" + } + ], + "title": "litellm_in_flight_requests", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Time (seconds) from request arrival at the proxy to the start of pre-call processing -- includes authentication and any ASGI-level queueing", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "s" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 12, + "y": 17 + }, + "id": 7, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "histogram_quantile(0.5, sum(rate(litellm_request_queue_time_seconds_bucket[$__rate_interval])) by (le))", + "legendFormat": "p50", + "range": true, + "refId": "A" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "histogram_quantile(0.95, sum(rate(litellm_request_queue_time_seconds_bucket[$__rate_interval])) by (le))", + "legendFormat": "p95", + "range": true, + "refId": "B" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "histogram_quantile(0.99, sum(rate(litellm_request_queue_time_seconds_bucket[$__rate_interval])) by (le))", + "legendFormat": "p99", + "range": true, + "refId": "C" + } + ], + "title": "litellm_request_queue_time_seconds p50 / p95 / p99", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Requests admitted by this worker (needs the admission control middleware enabled)", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "short" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 0, + "y": 25 + }, + "id": 8, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "litellm_admission_admitted_requests", + "legendFormat": "__auto", + "range": true, + "refId": "A" + } + ], + "title": "litellm_admission_admitted_requests", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Requests queued by this worker (needs the admission control middleware enabled)", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "short" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 12, + "y": 25 + }, + "id": 9, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "litellm_admission_queued_requests", + "legendFormat": "__auto", + "range": true, + "refId": "A" + } + ], + "title": "litellm_admission_queued_requests", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Requests rejected by this worker (needs the admission control middleware enabled)", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "short" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 0, + "y": 33 + }, + "id": 10, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_admission_rejected_requests_total[$__rate_interval])) by (reason)", + "legendFormat": "{{reason}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_admission_rejected_requests rate", + "type": "timeseries" + }, + { + "collapsed": false, + "gridPos": { + "h": 1, + "w": 24, + "x": 0, + "y": 41 + }, + "id": 11, + "panels": [], + "title": "Latency", + "type": "row" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "End-to-end latency (seconds) for a request to LiteLLM Proxy Server, from the moment the request reached the proxy through the end of processing -- includes authentication, pre-call hooks, the LLM API call, and post-call processing", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "s" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 0, + "y": 42 + }, + "id": 12, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "histogram_quantile(0.5, sum(rate(litellm_request_total_latency_metric_bucket[$__rate_interval])) by (le))", + "legendFormat": "p50", + "range": true, + "refId": "A" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "histogram_quantile(0.95, sum(rate(litellm_request_total_latency_metric_bucket[$__rate_interval])) by (le))", + "legendFormat": "p95", + "range": true, + "refId": "B" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "histogram_quantile(0.99, sum(rate(litellm_request_total_latency_metric_bucket[$__rate_interval])) by (le))", + "legendFormat": "p99", + "range": true, + "refId": "C" + } + ], + "title": "litellm_request_total_latency_metric p50 / p95 / p99", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Total latency (seconds) for a models LLM API call", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "s" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 12, + "y": 42 + }, + "id": 13, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "histogram_quantile(0.5, sum(rate(litellm_llm_api_latency_metric_bucket[$__rate_interval])) by (le))", + "legendFormat": "p50", + "range": true, + "refId": "A" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "histogram_quantile(0.95, sum(rate(litellm_llm_api_latency_metric_bucket[$__rate_interval])) by (le))", + "legendFormat": "p95", + "range": true, + "refId": "B" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "histogram_quantile(0.99, sum(rate(litellm_llm_api_latency_metric_bucket[$__rate_interval])) by (le))", + "legendFormat": "p99", + "range": true, + "refId": "C" + } + ], + "title": "litellm_llm_api_latency_metric p50 / p95 / p99", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Time to first token for a models LLM API call", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "s" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 0, + "y": 50 + }, + "id": 14, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "histogram_quantile(0.5, sum(rate(litellm_llm_api_time_to_first_token_metric_bucket[$__rate_interval])) by (le))", + "legendFormat": "p50", + "range": true, + "refId": "A" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "histogram_quantile(0.95, sum(rate(litellm_llm_api_time_to_first_token_metric_bucket[$__rate_interval])) by (le))", + "legendFormat": "p95", + "range": true, + "refId": "B" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "histogram_quantile(0.99, sum(rate(litellm_llm_api_time_to_first_token_metric_bucket[$__rate_interval])) by (le))", + "legendFormat": "p99", + "range": true, + "refId": "C" + } + ], + "title": "litellm_llm_api_time_to_first_token_metric p50 / p95 / p99", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Latency overhead (seconds) added by LiteLLM processing", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "s" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 12, + "y": 50 + }, + "id": 15, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "histogram_quantile(0.5, sum(rate(litellm_overhead_latency_metric_bucket[$__rate_interval])) by (le))", + "legendFormat": "p50", + "range": true, + "refId": "A" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "histogram_quantile(0.95, sum(rate(litellm_overhead_latency_metric_bucket[$__rate_interval])) by (le))", + "legendFormat": "p95", + "range": true, + "refId": "B" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "histogram_quantile(0.99, sum(rate(litellm_overhead_latency_metric_bucket[$__rate_interval])) by (le))", + "legendFormat": "p99", + "range": true, + "refId": "C" + } + ], + "title": "litellm_overhead_latency_metric p50 / p95 / p99", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Total internal latency (seconds) added by LiteLLM, including pre/post-call guardrails (excludes the LLM API call)", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "s" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 0, + "y": 58 + }, + "id": 16, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "histogram_quantile(0.5, sum(rate(litellm_overhead_with_guardrails_latency_metric_bucket[$__rate_interval])) by (le))", + "legendFormat": "p50", + "range": true, + "refId": "A" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "histogram_quantile(0.95, sum(rate(litellm_overhead_with_guardrails_latency_metric_bucket[$__rate_interval])) by (le))", + "legendFormat": "p95", + "range": true, + "refId": "B" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "histogram_quantile(0.99, sum(rate(litellm_overhead_with_guardrails_latency_metric_bucket[$__rate_interval])) by (le))", + "legendFormat": "p99", + "range": true, + "refId": "C" + } + ], + "title": "litellm_overhead_with_guardrails_latency_metric p50 / p95 / p99", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "LLM Deployment Analytics - Latency per output token", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "s" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 12, + "y": 58 + }, + "id": 17, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "histogram_quantile(0.5, sum(rate(litellm_deployment_latency_per_output_token_bucket[$__rate_interval])) by (le))", + "legendFormat": "p50", + "range": true, + "refId": "A" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "histogram_quantile(0.95, sum(rate(litellm_deployment_latency_per_output_token_bucket[$__rate_interval])) by (le))", + "legendFormat": "p95", + "range": true, + "refId": "B" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "histogram_quantile(0.99, sum(rate(litellm_deployment_latency_per_output_token_bucket[$__rate_interval])) by (le))", + "legendFormat": "p99", + "range": true, + "refId": "C" + } + ], + "title": "litellm_deployment_latency_per_output_token p50 / p95 / p99", + "type": "timeseries" + }, + { + "collapsed": false, + "gridPos": { + "h": 1, + "w": 24, + "x": 0, + "y": 66 + }, + "id": 18, + "panels": [], + "title": "Spend and tokens", + "type": "row" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Total spend on LLM requests", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "currencyUSD" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 0, + "y": 67 + }, + "id": 19, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_spend_metric_total[$__rate_interval])) by (model)", + "legendFormat": "{{model}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_spend_metric rate", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Total number of input + output tokens from LLM requests", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "short" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 12, + "y": 67 + }, + "id": 20, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_total_tokens_metric_total[$__rate_interval])) by (model)", + "legendFormat": "{{model}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_total_tokens_metric rate", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Total number of input tokens from LLM requests", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "short" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 0, + "y": 75 + }, + "id": 21, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_input_tokens_metric_total[$__rate_interval])) by (model)", + "legendFormat": "{{model}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_input_tokens_metric rate", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Total number of output tokens from LLM requests", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "short" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 12, + "y": 75 + }, + "id": 22, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_output_tokens_metric_total[$__rate_interval])) by (model)", + "legendFormat": "{{model}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_output_tokens_metric rate", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Provider-side cached input tokens (e.g. OpenAI prompt_tokens_details.cached_tokens, Anthropic cache_read_input_tokens)", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "short" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 0, + "y": 83 + }, + "id": 23, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_input_cached_tokens_metric_total[$__rate_interval])) by (model)", + "legendFormat": "{{model}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_input_cached_tokens_metric rate", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Provider-side input tokens written to prompt cache (e.g. Anthropic cache_creation_input_tokens)", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "short" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 12, + "y": 83 + }, + "id": 24, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_input_cache_creation_tokens_metric_total[$__rate_interval])) by (model)", + "legendFormat": "{{model}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_input_cache_creation_tokens_metric rate", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Audio input tokens reported in prompt_tokens_details.audio_tokens", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "short" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 0, + "y": 91 + }, + "id": 25, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_input_audio_tokens_metric_total[$__rate_interval])) by (model)", + "legendFormat": "{{model}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_input_audio_tokens_metric rate", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Audio output tokens reported in completion_tokens_details.audio_tokens", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "short" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 12, + "y": 91 + }, + "id": 26, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_output_audio_tokens_metric_total[$__rate_interval])) by (model)", + "legendFormat": "{{model}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_output_audio_tokens_metric rate", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Reasoning tokens reported in completion_tokens_details.reasoning_tokens", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "short" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 0, + "y": 99 + }, + "id": 27, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_output_reasoning_tokens_metric_total[$__rate_interval])) by (model)", + "legendFormat": "{{model}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_output_reasoning_tokens_metric rate", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Number of images generated, from the image generation response", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "short" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 12, + "y": 99 + }, + "id": 28, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_images_generated_metric_total[$__rate_interval])) by (model)", + "legendFormat": "{{model}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_images_generated_metric rate", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Seconds of video generated, from usage.duration_seconds on video generation calls", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "s" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 0, + "y": 107 + }, + "id": 29, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_video_duration_seconds_metric_total[$__rate_interval])) by (model)", + "legendFormat": "{{model}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_video_duration_seconds_metric rate", + "type": "timeseries" + }, + { + "collapsed": false, + "gridPos": { + "h": 1, + "w": 24, + "x": 0, + "y": 115 + }, + "id": 30, + "panels": [], + "title": "Cache", + "type": "row" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Total number of LiteLLM cache hits", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "short" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 0, + "y": 116 + }, + "id": 31, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_cache_hits_metric_total[$__rate_interval])) by (model)", + "legendFormat": "{{model}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_cache_hits_metric rate", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Total number of LiteLLM cache misses", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "short" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 12, + "y": 116 + }, + "id": 32, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_cache_misses_metric_total[$__rate_interval])) by (model)", + "legendFormat": "{{model}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_cache_misses_metric rate", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Total tokens served from LiteLLM cache", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "short" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 0, + "y": 124 + }, + "id": 33, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_cached_tokens_metric_total[$__rate_interval])) by (model)", + "legendFormat": "{{model}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_cached_tokens_metric rate", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Total prompt/input tokens read from provider prompt cache (e.g. OpenAI/Anthropic/Gemini/Bedrock)", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "short" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 12, + "y": 124 + }, + "id": 34, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_provider_cache_read_input_tokens_metric_total[$__rate_interval])) by (model)", + "legendFormat": "{{model}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_provider_cache_read_input_tokens_metric rate", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Total prompt/input tokens written to provider prompt cache (e.g. Anthropic/Bedrock)", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "short" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 0, + "y": 132 + }, + "id": 35, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_provider_cache_creation_input_tokens_metric_total[$__rate_interval])) by (model)", + "legendFormat": "{{model}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_provider_cache_creation_input_tokens_metric rate", + "type": "timeseries" + }, + { + "collapsed": false, + "gridPos": { + "h": 1, + "w": 24, + "x": 0, + "y": 140 + }, + "id": 36, + "panels": [], + "title": "LLM API deployments", + "type": "row" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "LLM Deployment Analytics - The state of the deployment: 0 = healthy, 1 = partial outage, 2 = complete outage", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "short" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 0, + "y": 141 + }, + "id": 37, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "max by (litellm_model_name, api_base) (litellm_deployment_state)", + "legendFormat": "{{litellm_model_name}} / {{api_base}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_deployment_state", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "LLM Deployment Analytics - Total number of LLM API calls via litellm - success + failure", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "short" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 12, + "y": 141 + }, + "id": 38, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_deployment_total_requests_total[$__rate_interval])) by (requested_model)", + "legendFormat": "{{requested_model}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_deployment_total_requests rate", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "LLM Deployment Analytics - Total number of successful LLM API calls via litellm", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "short" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 0, + "y": 149 + }, + "id": 39, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_deployment_success_responses_total[$__rate_interval])) by (requested_model)", + "legendFormat": "{{requested_model}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_deployment_success_responses rate", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "LLM Deployment Analytics - Total number of failed LLM API calls for a specific LLM deploymeny. exception_status is the status of the exception from the llm api", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "short" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 12, + "y": 149 + }, + "id": 40, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_deployment_failure_responses_total[$__rate_interval])) by (litellm_model_name, exception_class)", + "legendFormat": "{{litellm_model_name}} / {{exception_class}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_deployment_failure_responses rate", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "LLM Deployment Analytics - Number of times a deployment has been cooled down by LiteLLM load balancing logic. exception_status is the status of the exception that caused the deployment to be cooled down", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "short" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 0, + "y": 157 + }, + "id": 41, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_deployment_cooled_down_total[$__rate_interval])) by (litellm_model_name, exception_status)", + "legendFormat": "{{litellm_model_name}} / {{exception_status}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_deployment_cooled_down rate", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "LLM Deployment Analytics - Number of successful fallback requests from primary model -> fallback model", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "short" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 12, + "y": 157 + }, + "id": 42, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_deployment_successful_fallbacks_total[$__rate_interval])) by (requested_model, fallback_model)", + "legendFormat": "{{requested_model}} / {{fallback_model}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_deployment_successful_fallbacks rate", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "LLM Deployment Analytics - Number of failed fallback requests from primary model -> fallback model", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "short" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 0, + "y": 165 + }, + "id": 43, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_deployment_failed_fallbacks_total[$__rate_interval])) by (requested_model, fallback_model)", + "legendFormat": "{{requested_model}} / {{fallback_model}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_deployment_failed_fallbacks rate", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Deployment RPM limit found in config", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "short" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 12, + "y": 165 + }, + "id": 44, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "max by (litellm_model_name, api_base) (litellm_deployment_rpm_limit)", + "legendFormat": "{{litellm_model_name}} / {{api_base}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_deployment_rpm_limit", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Deployment TPM limit found in config", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "short" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 0, + "y": 173 + }, + "id": 45, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "max by (litellm_model_name, api_base) (litellm_deployment_tpm_limit)", + "legendFormat": "{{litellm_model_name}} / {{api_base}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_deployment_tpm_limit", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "LLM Deployment Analytics - remaining requests for model, returned from LLM API Provider", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "short" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 12, + "y": 173 + }, + "id": 46, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "min by (model_group, api_provider) (litellm_remaining_requests_metric)", + "legendFormat": "{{model_group}} / {{api_provider}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_remaining_requests_metric", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "remaining tokens for model, returned from LLM API Provider", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "short" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 0, + "y": 181 + }, + "id": 47, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "min by (model_group, api_provider) (litellm_remaining_tokens_metric)", + "legendFormat": "{{model_group}} / {{api_provider}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_remaining_tokens_metric", + "type": "timeseries" + }, + { + "collapsed": false, + "gridPos": { + "h": 1, + "w": 24, + "x": 0, + "y": 189 + }, + "id": 48, + "panels": [], + "title": "Key and team rate limits", + "type": "row" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Remaining Requests API Key can make for model (model based rpm limit on key)", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "short" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 0, + "y": 190 + }, + "id": 49, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "min by (api_key_alias, model) (litellm_remaining_api_key_requests_for_model)", + "legendFormat": "{{api_key_alias}} / {{model}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_remaining_api_key_requests_for_model", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Remaining Tokens API Key can make for model (model based tpm limit on key)", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "short" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 12, + "y": 190 + }, + "id": 50, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "min by (api_key_alias, model) (litellm_remaining_api_key_tokens_for_model)", + "legendFormat": "{{api_key_alias}} / {{model}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_remaining_api_key_tokens_for_model", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Configured rate limit for the API Key in the current window (rpm_limit / tpm_limit), by rate_limit_type", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "short" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 0, + "y": 198 + }, + "id": 51, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "max by (api_key_alias, rate_limit_type) (litellm_api_key_rate_limit_allowed_metric)", + "legendFormat": "{{api_key_alias}} / {{rate_limit_type}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_api_key_rate_limit_allowed_metric", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Requests or tokens the API Key has consumed in the current rate limit window, by rate_limit_type", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "short" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 12, + "y": 198 + }, + "id": 52, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "max by (api_key_alias, rate_limit_type) (litellm_api_key_rate_limit_used_metric)", + "legendFormat": "{{api_key_alias}} / {{rate_limit_type}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_api_key_rate_limit_used_metric", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Configured rate limit for the Team in the current window (team rpm_limit / tpm_limit), by rate_limit_type", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "short" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 0, + "y": 206 + }, + "id": 53, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "max by (team_alias, rate_limit_type) (litellm_team_rate_limit_allowed_metric)", + "legendFormat": "{{team_alias}} / {{rate_limit_type}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_team_rate_limit_allowed_metric", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Requests or tokens the Team has consumed in the current rate limit window, by rate_limit_type", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "short" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 12, + "y": 206 + }, + "id": 54, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "max by (team_alias, rate_limit_type) (litellm_team_rate_limit_used_metric)", + "legendFormat": "{{team_alias}} / {{rate_limit_type}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_team_rate_limit_used_metric", + "type": "timeseries" + }, + { + "collapsed": false, + "gridPos": { + "h": 1, + "w": 24, + "x": 0, + "y": 214 + }, + "id": 55, + "panels": [], + "title": "Budgets", + "type": "row" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Remaining budget for team", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "currencyUSD" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 0, + "y": 215 + }, + "id": 56, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "min by (team_alias) (litellm_remaining_team_budget_metric)", + "legendFormat": "{{team_alias}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_remaining_team_budget_metric", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Maximum budget set for team", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "currencyUSD" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 12, + "y": 215 + }, + "id": 57, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "max by (team_alias) (litellm_team_max_budget_metric)", + "legendFormat": "{{team_alias}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_team_max_budget_metric", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Remaining days for team budget to be reset", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "h" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 0, + "y": 223 + }, + "id": 58, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "max by (team_alias) (litellm_team_budget_remaining_hours_metric)", + "legendFormat": "{{team_alias}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_team_budget_remaining_hours_metric", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Remaining budget for api key", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "currencyUSD" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 12, + "y": 223 + }, + "id": 59, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "min by (api_key_alias) (litellm_remaining_api_key_budget_metric)", + "legendFormat": "{{api_key_alias}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_remaining_api_key_budget_metric", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Maximum budget set for api key", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "currencyUSD" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 0, + "y": 231 + }, + "id": 60, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "max by (api_key_alias) (litellm_api_key_max_budget_metric)", + "legendFormat": "{{api_key_alias}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_api_key_max_budget_metric", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Remaining hours for api key budget to be reset", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "h" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 12, + "y": 231 + }, + "id": 61, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "max by (api_key_alias) (litellm_api_key_budget_remaining_hours_metric)", + "legendFormat": "{{api_key_alias}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_api_key_budget_remaining_hours_metric", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Remaining budget for user", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "currencyUSD" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 0, + "y": 239 + }, + "id": 62, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "min by (user) (litellm_remaining_user_budget_metric)", + "legendFormat": "{{user}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_remaining_user_budget_metric", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Maximum budget set for user", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "currencyUSD" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 12, + "y": 239 + }, + "id": 63, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "max by (user) (litellm_user_max_budget_metric)", + "legendFormat": "{{user}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_user_max_budget_metric", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Remaining hours for user budget to be reset", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "h" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 0, + "y": 247 + }, + "id": 64, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "max by (user) (litellm_user_budget_remaining_hours_metric)", + "legendFormat": "{{user}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_user_budget_remaining_hours_metric", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Remaining budget for org", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "currencyUSD" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 12, + "y": 247 + }, + "id": 65, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "min by (org_alias) (litellm_remaining_org_budget_metric)", + "legendFormat": "{{org_alias}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_remaining_org_budget_metric", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Maximum budget set for org", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "currencyUSD" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 0, + "y": 255 + }, + "id": 66, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "max by (org_alias) (litellm_org_max_budget_metric)", + "legendFormat": "{{org_alias}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_org_max_budget_metric", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Remaining hours for org budget to be reset", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "h" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 12, + "y": 255 + }, + "id": 67, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "max by (org_alias) (litellm_org_budget_remaining_hours_metric)", + "legendFormat": "{{org_alias}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_org_budget_remaining_hours_metric", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Remaining budget for customer (end user)", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "currencyUSD" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 0, + "y": 263 + }, + "id": 68, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "min by (end_user) (litellm_remaining_customer_budget_metric)", + "legendFormat": "{{end_user}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_remaining_customer_budget_metric", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Maximum budget set for customer (end user)", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "currencyUSD" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 12, + "y": 263 + }, + "id": 69, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "max by (end_user) (litellm_customer_max_budget_metric)", + "legendFormat": "{{end_user}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_customer_max_budget_metric", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Remaining hours for customer (end user) budget to be reset", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "h" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 0, + "y": 271 + }, + "id": 70, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "max by (end_user) (litellm_customer_budget_remaining_hours_metric)", + "legendFormat": "{{end_user}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_customer_budget_remaining_hours_metric", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Remaining budget for provider - used when you set provider budget limits", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "currencyUSD" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 12, + "y": 271 + }, + "id": 71, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "min by (api_provider) (litellm_provider_remaining_budget_metric)", + "legendFormat": "{{api_provider}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_provider_remaining_budget_metric", + "type": "timeseries" + }, + { + "collapsed": false, + "gridPos": { + "h": 1, + "w": 24, + "x": 0, + "y": 279 + }, + "id": 72, + "panels": [], + "title": "Guardrails", + "type": "row" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Total number of guardrail invocations", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "short" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 0, + "y": 280 + }, + "id": 73, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_guardrail_requests_total[$__rate_interval])) by (guardrail_name, status)", + "legendFormat": "{{guardrail_name}} / {{status}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_guardrail_requests rate", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Total number of errors encountered during guardrail execution", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "short" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 12, + "y": 280 + }, + "id": 74, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_guardrail_errors_total[$__rate_interval])) by (guardrail_name, error_type)", + "legendFormat": "{{guardrail_name}} / {{error_type}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_guardrail_errors rate", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Latency (seconds) for guardrail execution", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "s" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 0, + "y": 288 + }, + "id": 75, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "histogram_quantile(0.5, sum(rate(litellm_guardrail_latency_seconds_bucket[$__rate_interval])) by (le))", + "legendFormat": "p50", + "range": true, + "refId": "A" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "histogram_quantile(0.95, sum(rate(litellm_guardrail_latency_seconds_bucket[$__rate_interval])) by (le))", + "legendFormat": "p95", + "range": true, + "refId": "B" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "histogram_quantile(0.99, sum(rate(litellm_guardrail_latency_seconds_bucket[$__rate_interval])) by (le))", + "legendFormat": "p99", + "range": true, + "refId": "C" + } + ], + "title": "litellm_guardrail_latency_seconds p50 / p95 / p99", + "type": "timeseries" + }, + { + "collapsed": false, + "gridPos": { + "h": 1, + "w": 24, + "x": 0, + "y": 296 + }, + "id": 76, + "panels": [], + "title": "MCP", + "type": "row" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Total MCP tool calls, segmented by tool and server name", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "short" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 0, + "y": 297 + }, + "id": 77, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_mcp_tool_calls_total[$__rate_interval])) by (mcp_server_name, mcp_tool_name)", + "legendFormat": "{{mcp_server_name}} / {{mcp_tool_name}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_mcp_tool_calls rate", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Total spend on MCP tool calls, segmented by tool and server name", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "currencyUSD" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 12, + "y": 297 + }, + "id": 78, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_mcp_tool_call_spend_metric_total[$__rate_interval])) by (mcp_server_name, mcp_tool_name)", + "legendFormat": "{{mcp_server_name}} / {{mcp_tool_name}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_mcp_tool_call_spend_metric rate", + "type": "timeseries" + }, + { + "collapsed": false, + "gridPos": { + "h": 1, + "w": 24, + "x": 0, + "y": 305 + }, + "id": 79, + "panels": [], + "title": "Managed files and batches", + "type": "row" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Total number of managed files created", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "short" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 0, + "y": 306 + }, + "id": 80, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_managed_file_created_total[$__rate_interval])) by (model)", + "legendFormat": "{{model}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_managed_file_created rate", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Total number of managed file deletions (success or blocked)", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "short" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 12, + "y": 306 + }, + "id": 81, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_managed_file_deleted_total[$__rate_interval])) by (result)", + "legendFormat": "{{result}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_managed_file_deleted rate", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Size of the most recent managed batch file in bytes (last-seen value per label combination)", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "bytes" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 0, + "y": 314 + }, + "id": 82, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "max by (purpose, model) (litellm_managed_file_size_bytes)", + "legendFormat": "{{purpose}} / {{model}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_managed_file_size_bytes", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Total number of managed batches created", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "short" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 12, + "y": 314 + }, + "id": 83, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_managed_batch_created_total[$__rate_interval])) by (model)", + "legendFormat": "{{model}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_managed_batch_created rate", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Duration of completed managed batches in seconds (completed_at - created_at)", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "s" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 0, + "y": 322 + }, + "id": 84, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "histogram_quantile(0.5, sum(rate(litellm_managed_batch_duration_seconds_bucket[$__rate_interval])) by (le))", + "legendFormat": "p50", + "range": true, + "refId": "A" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "histogram_quantile(0.95, sum(rate(litellm_managed_batch_duration_seconds_bucket[$__rate_interval])) by (le))", + "legendFormat": "p95", + "range": true, + "refId": "B" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "histogram_quantile(0.99, sum(rate(litellm_managed_batch_duration_seconds_bucket[$__rate_interval])) by (le))", + "legendFormat": "p99", + "range": true, + "refId": "C" + } + ], + "title": "litellm_managed_batch_duration_seconds p50 / p95 / p99", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Number of unprocessed batches found by the last CheckBatchCost poll", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "short" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 12, + "y": 322 + }, + "id": 85, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "litellm_check_batch_cost_jobs_polled", + "legendFormat": "__auto", + "range": true, + "refId": "A" + } + ], + "title": "litellm_check_batch_cost_jobs_polled", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Total number of batches successfully cost-tracked by CheckBatchCost", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "short" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 0, + "y": 330 + }, + "id": 86, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_check_batch_cost_jobs_processed_total[$__rate_interval])) by (model, api_provider)", + "legendFormat": "{{model}} / {{api_provider}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_check_batch_cost_jobs_processed rate", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Total number of errors in CheckBatchCost by error type", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "short" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 12, + "y": 330 + }, + "id": 87, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_check_batch_cost_errors_total[$__rate_interval])) by (error_type)", + "legendFormat": "{{error_type}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_check_batch_cost_errors rate", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Unix timestamp of the last CheckBatchCost job run", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "s" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 0, + "y": 338 + }, + "id": 88, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "time() - (litellm_check_batch_cost_last_run_timestamp > 0)", + "legendFormat": "seconds since last run", + "range": true, + "refId": "A" + } + ], + "title": "litellm_check_batch_cost_last_run_timestamp (seconds since last run)", + "type": "timeseries" + }, + { + "collapsed": false, + "gridPos": { + "h": 1, + "w": 24, + "x": 0, + "y": 346 + }, + "id": 89, + "panels": [], + "title": "Users, teams and callbacks", + "type": "row" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Total number of users in LiteLLM", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "short" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 0, + "y": 347 + }, + "id": 90, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "litellm_total_users", + "legendFormat": "__auto", + "range": true, + "refId": "A" + } + ], + "title": "litellm_total_users", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Number of billable users in LiteLLM (excludes SCIM-deactivated users)", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "short" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 12, + "y": 347 + }, + "id": 91, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "litellm_active_users", + "legendFormat": "__auto", + "range": true, + "refId": "A" + } + ], + "title": "litellm_active_users", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Total number of teams in LiteLLM", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "short" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 0, + "y": 355 + }, + "id": 92, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "litellm_teams_count", + "legendFormat": "__auto", + "range": true, + "refId": "A" + } + ], + "title": "litellm_teams_count", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Number of members in a team", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "short" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 12, + "y": 355 + }, + "id": 93, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "max by (team_alias) (litellm_team_members_metric)", + "legendFormat": "{{team_alias}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_team_members_metric", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Total number of failures when emitting logs to callbacks (e.g. s3_v2, langfuse, etc)", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "short" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 0, + "y": 363 + }, + "id": 94, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_callback_logging_failures_metric_total[$__rate_interval])) by (callback_name)", + "legendFormat": "{{callback_name}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_callback_logging_failures_metric rate", + "type": "timeseries" + }, + { + "collapsed": false, + "gridPos": { + "h": 1, + "w": 24, + "x": 0, + "y": 371 + }, + "id": 95, + "panels": [], + "title": "Redis circuit breaker (needs a Redis cache)", + "type": "row" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Number of Redis circuit breakers currently in each state", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "short" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 0, + "y": 372 + }, + "id": 96, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum by (state) (litellm_redis_circuit_breaker_state)", + "legendFormat": "{{state}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_redis_circuit_breaker_state", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Redis circuit breaker state transitions", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "short" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 12, + "y": 372 + }, + "id": 97, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_redis_circuit_breaker_transitions_total[$__rate_interval])) by (state)", + "legendFormat": "{{state}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_redis_circuit_breaker_transitions rate", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Redis health failures counted by the circuit breaker", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "short" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 0, + "y": 380 + }, + "id": 98, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_redis_circuit_breaker_failures_total[$__rate_interval])) by (failure_class)", + "legendFormat": "{{failure_class}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_redis_circuit_breaker_failures rate", + "type": "timeseries" + }, + { + "collapsed": false, + "gridPos": { + "h": 1, + "w": 24, + "x": 0, + "y": 388 + }, + "id": 99, + "panels": [], + "title": "Spend log cleanup job (needs spend log retention enabled)", + "type": "row" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Retention cleanup runs, labelled by why the run ended", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "short" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 0, + "y": 389 + }, + "id": 100, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_spend_log_cleanup_runs_total[$__rate_interval])) by (outcome)", + "legendFormat": "{{outcome}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_spend_log_cleanup_runs rate", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Rows deleted by the spend-log retention cleanup job", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "short" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 12, + "y": 389 + }, + "id": 101, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_spend_log_cleanup_rows_deleted_total[$__rate_interval])) by (table)", + "legendFormat": "{{table}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_spend_log_cleanup_rows_deleted rate", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Expired rows still awaiting deletion, counted only up to SPEND_LOG_CLEANUP_REMAINING_COUNT_CAP so the probe itself cannot scan a large table; a value equal to that cap means at least that many remain", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "short" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 0, + "y": 397 + }, + "id": 102, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "max by (table) (litellm_spend_log_cleanup_rows_remaining)", + "legendFormat": "{{table}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_spend_log_cleanup_rows_remaining", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Wall-clock duration of one retention cleanup delete batch", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "s" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 12, + "y": 397 + }, + "id": 103, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "histogram_quantile(0.5, sum(rate(litellm_spend_log_cleanup_batch_duration_seconds_bucket[$__rate_interval])) by (le))", + "legendFormat": "p50", + "range": true, + "refId": "A" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "histogram_quantile(0.95, sum(rate(litellm_spend_log_cleanup_batch_duration_seconds_bucket[$__rate_interval])) by (le))", + "legendFormat": "p95", + "range": true, + "refId": "B" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "histogram_quantile(0.99, sum(rate(litellm_spend_log_cleanup_batch_duration_seconds_bucket[$__rate_interval])) by (le))", + "legendFormat": "p99", + "range": true, + "refId": "C" + } + ], + "title": "litellm_spend_log_cleanup_batch_duration_seconds p50 / p95 / p99", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Retention cleanup delete batches that raised", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "short" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 0, + "y": 405 + }, + "id": 104, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_spend_log_cleanup_batch_failures_total[$__rate_interval])) by (table)", + "legendFormat": "{{table}}", + "range": true, + "refId": "A" + } + ], + "title": "litellm_spend_log_cleanup_batch_failures rate", + "type": "timeseries" + }, + { + "collapsed": false, + "gridPos": { + "h": 1, + "w": 24, + "x": 0, + "y": 413 + }, + "id": 105, + "panels": [], + "title": "Service callbacks (needs service_callback: prometheus_system)", + "type": "row" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "p95 latency per internal service: redis, postgres, router, auth, batch writes, budget reset, proxy pre-call hooks and the proxy itself (self)", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "s" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 0, + "y": 414 + }, + "id": 106, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "histogram_quantile(0.95, sum(rate(litellm_auth_latency_bucket[$__rate_interval])) by (le))", + "legendFormat": "auth", + "range": true, + "refId": "A" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "histogram_quantile(0.95, sum(rate(litellm_batch_write_to_db_latency_bucket[$__rate_interval])) by (le))", + "legendFormat": "batch_write_to_db", + "range": true, + "refId": "B" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "histogram_quantile(0.95, sum(rate(litellm_postgres_latency_bucket[$__rate_interval])) by (le))", + "legendFormat": "postgres", + "range": true, + "refId": "C" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "histogram_quantile(0.95, sum(rate(litellm_proxy_pre_call_latency_bucket[$__rate_interval])) by (le))", + "legendFormat": "proxy_pre_call", + "range": true, + "refId": "D" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "histogram_quantile(0.95, sum(rate(litellm_redis_latency_bucket[$__rate_interval])) by (le))", + "legendFormat": "redis", + "range": true, + "refId": "E" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "histogram_quantile(0.95, sum(rate(litellm_redis_daily_org_spend_update_queue_latency_bucket[$__rate_interval])) by (le))", + "legendFormat": "redis_daily_org_spend_update_queue", + "range": true, + "refId": "F" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "histogram_quantile(0.95, sum(rate(litellm_redis_daily_tag_spend_update_queue_latency_bucket[$__rate_interval])) by (le))", + "legendFormat": "redis_daily_tag_spend_update_queue", + "range": true, + "refId": "G" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "histogram_quantile(0.95, sum(rate(litellm_redis_daily_team_spend_update_queue_latency_bucket[$__rate_interval])) by (le))", + "legendFormat": "redis_daily_team_spend_update_queue", + "range": true, + "refId": "H" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "histogram_quantile(0.95, sum(rate(litellm_redis_window_spend_update_queue_latency_bucket[$__rate_interval])) by (le))", + "legendFormat": "redis_window_spend_update_queue", + "range": true, + "refId": "I" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "histogram_quantile(0.95, sum(rate(litellm_reset_budget_job_latency_bucket[$__rate_interval])) by (le))", + "legendFormat": "reset_budget_job", + "range": true, + "refId": "J" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "histogram_quantile(0.95, sum(rate(litellm_router_latency_bucket[$__rate_interval])) by (le))", + "legendFormat": "router", + "range": true, + "refId": "K" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "histogram_quantile(0.95, sum(rate(litellm_self_latency_bucket[$__rate_interval])) by (le))", + "legendFormat": "self", + "range": true, + "refId": "L" + } + ], + "title": "Service latency p95 (litellm__latency)", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Requests per second handled by each internal service", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "reqps" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 12, + "y": 414 + }, + "id": 107, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_auth_total_requests_total[$__rate_interval]))", + "legendFormat": "auth", + "range": true, + "refId": "A" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_batch_write_to_db_total_requests_total[$__rate_interval]))", + "legendFormat": "batch_write_to_db", + "range": true, + "refId": "B" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_postgres_total_requests_total[$__rate_interval]))", + "legendFormat": "postgres", + "range": true, + "refId": "C" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_proxy_pre_call_total_requests_total[$__rate_interval]))", + "legendFormat": "proxy_pre_call", + "range": true, + "refId": "D" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_redis_total_requests_total[$__rate_interval]))", + "legendFormat": "redis", + "range": true, + "refId": "E" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_redis_daily_org_spend_update_queue_total_requests_total[$__rate_interval]))", + "legendFormat": "redis_daily_org_spend_update_queue", + "range": true, + "refId": "F" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_redis_daily_tag_spend_update_queue_total_requests_total[$__rate_interval]))", + "legendFormat": "redis_daily_tag_spend_update_queue", + "range": true, + "refId": "G" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_redis_daily_team_spend_update_queue_total_requests_total[$__rate_interval]))", + "legendFormat": "redis_daily_team_spend_update_queue", + "range": true, + "refId": "H" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_redis_window_spend_update_queue_total_requests_total[$__rate_interval]))", + "legendFormat": "redis_window_spend_update_queue", + "range": true, + "refId": "I" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_reset_budget_job_total_requests_total[$__rate_interval]))", + "legendFormat": "reset_budget_job", + "range": true, + "refId": "J" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_router_total_requests_total[$__rate_interval]))", + "legendFormat": "router", + "range": true, + "refId": "K" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_self_total_requests_total[$__rate_interval]))", + "legendFormat": "self", + "range": true, + "refId": "L" + } + ], + "title": "Service request rate (litellm__total_requests)", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Failed requests per second per internal service, split by exception class", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "reqps" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 0, + "y": 422 + }, + "id": 108, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_auth_failed_requests_total[$__rate_interval])) by (error_class)", + "legendFormat": "auth / {{error_class}}", + "range": true, + "refId": "A" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_batch_write_to_db_failed_requests_total[$__rate_interval])) by (error_class)", + "legendFormat": "batch_write_to_db / {{error_class}}", + "range": true, + "refId": "B" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_postgres_failed_requests_total[$__rate_interval])) by (error_class)", + "legendFormat": "postgres / {{error_class}}", + "range": true, + "refId": "C" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_proxy_pre_call_failed_requests_total[$__rate_interval])) by (error_class)", + "legendFormat": "proxy_pre_call / {{error_class}}", + "range": true, + "refId": "D" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_redis_failed_requests_total[$__rate_interval])) by (error_class)", + "legendFormat": "redis / {{error_class}}", + "range": true, + "refId": "E" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_redis_daily_org_spend_update_queue_failed_requests_total[$__rate_interval])) by (error_class)", + "legendFormat": "redis_daily_org_spend_update_queue / {{error_class}}", + "range": true, + "refId": "F" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_redis_daily_tag_spend_update_queue_failed_requests_total[$__rate_interval])) by (error_class)", + "legendFormat": "redis_daily_tag_spend_update_queue / {{error_class}}", + "range": true, + "refId": "G" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_redis_daily_team_spend_update_queue_failed_requests_total[$__rate_interval])) by (error_class)", + "legendFormat": "redis_daily_team_spend_update_queue / {{error_class}}", + "range": true, + "refId": "H" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_redis_window_spend_update_queue_failed_requests_total[$__rate_interval])) by (error_class)", + "legendFormat": "redis_window_spend_update_queue / {{error_class}}", + "range": true, + "refId": "I" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_reset_budget_job_failed_requests_total[$__rate_interval])) by (error_class)", + "legendFormat": "reset_budget_job / {{error_class}}", + "range": true, + "refId": "J" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_router_failed_requests_total[$__rate_interval])) by (error_class)", + "legendFormat": "router / {{error_class}}", + "range": true, + "refId": "K" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "sum(rate(litellm_self_failed_requests_total[$__rate_interval])) by (error_class)", + "legendFormat": "self / {{error_class}}", + "range": true, + "refId": "L" + } + ], + "title": "Service failure rate (litellm__failed_requests)", + "type": "timeseries" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "description": "Items waiting in the in-memory and Redis spend update queues plus the pod lock manager", + "fieldConfig": { + "defaults": { + "color": { + "mode": "palette-classic" + }, + "custom": { + "drawStyle": "line", + "fillOpacity": 10, + "lineWidth": 1, + "showPoints": "never", + "spanNulls": false + }, + "unit": "short" + }, + "overrides": [] + }, + "gridPos": { + "h": 8, + "w": 12, + "x": 12, + "y": 422 + }, + "id": 109, + "options": { + "legend": { + "calcs": [], + "displayMode": "list", + "placement": "bottom", + "showLegend": true + }, + "tooltip": { + "mode": "multi", + "sort": "desc" + } + }, + "targets": [ + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "max(litellm_in_memory_daily_spend_update_queue_size)", + "legendFormat": "in_memory_daily_spend_update_queue", + "range": true, + "refId": "A" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "max(litellm_in_memory_spend_update_queue_size)", + "legendFormat": "in_memory_spend_update_queue", + "range": true, + "refId": "B" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "max(litellm_pod_lock_manager_size)", + "legendFormat": "pod_lock_manager", + "range": true, + "refId": "C" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "max(litellm_redis_daily_agent_spend_update_queue_size)", + "legendFormat": "redis_daily_agent_spend_update_queue", + "range": true, + "refId": "D" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "max(litellm_redis_daily_end_user_spend_update_queue_size)", + "legendFormat": "redis_daily_end_user_spend_update_queue", + "range": true, + "refId": "E" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "max(litellm_redis_daily_spend_update_queue_size)", + "legendFormat": "redis_daily_spend_update_queue", + "range": true, + "refId": "F" + }, + { + "datasource": { + "type": "prometheus", + "uid": "${DS_PROMETHEUS}" + }, + "editorMode": "code", + "expr": "max(litellm_redis_spend_update_queue_size)", + "legendFormat": "redis_spend_update_queue", + "range": true, + "refId": "G" + } + ], + "title": "Spend update queue sizes (litellm__size)", + "type": "timeseries" + } + ], + "preload": false, + "refresh": "30s", + "schemaVersion": 40, + "tags": [ + "litellm", + "prometheus" + ], + "templating": { + "list": [ + { + "current": { + "selected": false, + "text": "prometheus", + "value": "prometheus" + }, + "hide": 0, + "includeAll": false, + "label": "datasource", + "multi": false, + "name": "DS_PROMETHEUS", + "options": [], + "query": "prometheus", + "queryValue": "", + "refresh": 1, + "regex": "", + "skipUrlSync": false, + "type": "datasource" + } + ] + }, + "time": { + "from": "now-6h", + "to": "now" + }, + "timepicker": {}, + "timezone": "browser", + "title": "LiteLLM All Prometheus Metrics", + "uid": "litellm-all-prometheus-metrics", + "version": 1, + "weekStart": "" +} diff --git a/cookbook/litellm_proxy_server/grafana_dashboard/dashboard_all_metrics/readme.md b/cookbook/litellm_proxy_server/grafana_dashboard/dashboard_all_metrics/readme.md new file mode 100644 index 00000000000..6c491153562 --- /dev/null +++ b/cookbook/litellm_proxy_server/grafana_dashboard/dashboard_all_metrics/readme.md @@ -0,0 +1,11 @@ +# LiteLLM All Prometheus Metrics dashboard + +Every `litellm_*` metric family the proxy can expose on `/metrics` (134 families across 95 panels), grouped into rows: proxy traffic, latency, spend and tokens, cache, LLM API deployments, key and team rate limits, budgets, guardrails, MCP, managed files and batches, users and teams, the Redis circuit breaker, the spend log cleanup job, and the `prometheus_system` service callback metrics (per-service latency, request and failure rates, spend update queue sizes). Panel titles are the metric names so you can grep the JSON for the metric you care about + +Import `grafana_dashboard.json` from **Dashboards > New > Import** and pick your Prometheus data source when prompted (the `DS_PROMETHEUS` variable). Counters are plotted as `rate()` over `$__rate_interval`, histograms as p50 / p95 / p99, gauges as the raw value grouped by the most useful label. Every query names the metric exactly as the proxy emits it (counters carry the `_total` suffix the Prometheus client adds), and `tests/test_litellm/integrations/test_prometheus_metric_name_consistency.py` fails if a metric is renamed without updating this dashboard + +The first eleven rows need only `callbacks: ["prometheus"]`. The last three rows and the `litellm_admission_*` panels are emitted by other subsystems and stay empty until those are on: the service callback row needs `service_callback: ["prometheus_system"]` in `litellm_settings`, the circuit breaker row needs a Redis cache, the cleanup row needs spend log retention, and admission control needs its middleware enabled. Within the base rows, many panels only fill in once the matching feature is in use: budgets need keys, teams, users or orgs with `max_budget` set, cache panels need caching on, guardrail and MCP panels need those features configured, deployment health needs the router with more than one deployment or a failure to record, and `litellm_in_flight_requests` needs traffic at scrape time. An empty panel for a feature you do not use is expected + +## Pre-requisites + +Prometheus metrics on the proxy: https://docs.litellm.ai/docs/proxy/prometheus diff --git a/cookbook/litellm_proxy_server/grafana_dashboard/dashboard_v2/grafana_dashboard.json b/cookbook/litellm_proxy_server/grafana_dashboard/dashboard_v2/grafana_dashboard.json index 503364d8ff2..7a08cd5c5e9 100644 --- a/cookbook/litellm_proxy_server/grafana_dashboard/dashboard_v2/grafana_dashboard.json +++ b/cookbook/litellm_proxy_server/grafana_dashboard/dashboard_v2/grafana_dashboard.json @@ -476,7 +476,7 @@ "uid": "${DS_PROMETHEUS}" }, "editorMode": "code", - "expr": "topk(5, sort(litellm_remaining_requests))", + "expr": "topk(5, sort(litellm_remaining_requests_metric))", "legendFormat": "__auto", "range": true, "refId": "A" @@ -573,7 +573,7 @@ "uid": "${DS_PROMETHEUS}" }, "editorMode": "code", - "expr": "topk(5, sort(litellm_remaining_tokens))", + "expr": "topk(5, sort(litellm_remaining_tokens_metric))", "legendFormat": "__auto", "range": true, "refId": "A" diff --git a/cookbook/litellm_proxy_server/grafana_dashboard/readme.md b/cookbook/litellm_proxy_server/grafana_dashboard/readme.md index a1564a406e0..f10235f0073 100644 --- a/cookbook/litellm_proxy_server/grafana_dashboard/readme.md +++ b/cookbook/litellm_proxy_server/grafana_dashboard/readme.md @@ -6,8 +6,14 @@ This folder contains the `json` for creating Grafana Dashboards Charts the `gen_ai.*` metrics from the OpenTelemetry v2 integration: spend, tokens, request rate, and latency percentiles by model. Separate from the dashboards below, which chart the `litellm_*` Prometheus metrics. +## [LiteLLM All Prometheus Metrics dashboard](./dashboard_all_metrics) + +Every `litellm_*` Prometheus metric family the proxy can emit (134 families, 95 panels) grouped by theme: traffic, latency, spend and tokens, cache, deployments, rate limits, budgets, guardrails, MCP, managed files and batches, users and teams, plus the Redis circuit breaker, spend log cleanup and `prometheus_system` service metrics. Start here if you want everything on one screen; see its [readme](./dashboard_all_metrics/readme.md) for import steps and which panels need a feature enabled before they show data + ## [LiteLLM v2 Dashboard](./dashboard_v2) +A compact view of proxy request rate, failures, latency and the top remaining-request / remaining-token gauges per model group + grafana_1 grafana_2 grafana_3 diff --git a/gateway/routes/allowlist.py b/gateway/routes/allowlist.py index 099c6d5179f..915ce1af219 100644 --- a/gateway/routes/allowlist.py +++ b/gateway/routes/allowlist.py @@ -96,6 +96,7 @@ GATEWAY_PATH_PREFIXES: tuple[str, ...] = ( "/langfuse/", "/vllm/", "/mistral/", + "/typesafe/", "/nvidia_nim/", "/groq/", "/voyage/", diff --git a/litellm-rust/Cargo.lock b/litellm-rust/Cargo.lock index cf8a2442397..88dc837dd95 100644 --- a/litellm-rust/Cargo.lock +++ b/litellm-rust/Cargo.lock @@ -2016,6 +2016,7 @@ dependencies = [ "litellm-auth-azure", "litellm-auth-gcp", "litellm-framing", + "litellm-providers", "mime_guess", "moka", "rand 0.8.7", @@ -2052,6 +2053,18 @@ dependencies = [ "tokio", ] +[[package]] +name = "litellm-providers" +version = "0.1.0" +dependencies = [ + "litellm-auth", + "litellm-auth-aws", + "rstest", + "serde", + "serde_json", + "thiserror 2.0.19", +] + [[package]] name = "litellm-python-bridge" version = "0.1.0" diff --git a/litellm-rust/Cargo.toml b/litellm-rust/Cargo.toml index a63277d2e23..33cbd4f8b12 100644 --- a/litellm-rust/Cargo.toml +++ b/litellm-rust/Cargo.toml @@ -16,6 +16,7 @@ litellm-auth = { path = "crates/auth" } litellm-auth-aws = { path = "crates/auth-aws" } litellm-auth-azure = { path = "crates/auth-azure" } litellm-auth-gcp = { path = "crates/auth-gcp" } +litellm-providers = { path = "crates/providers" } litellm-cache = { path = "crates/cache" } litellm-cache-memory = { path = "crates/cache-memory" } litellm-token-counter = { path = "crates/token-counter" } diff --git a/litellm-rust/crates/core/Cargo.toml b/litellm-rust/crates/core/Cargo.toml index 6eacfad9fe7..7a836b4c95a 100644 --- a/litellm-rust/crates/core/Cargo.toml +++ b/litellm-rust/crates/core/Cargo.toml @@ -15,6 +15,7 @@ litellm-auth.workspace = true litellm-auth-aws.workspace = true litellm-auth-azure.workspace = true litellm-auth-gcp.workspace = true +litellm-providers.workspace = true litellm-framing.workspace = true moka.workspace = true mime_guess = "2.0.5" diff --git a/litellm-rust/crates/core/src/audio_transcription/error.rs b/litellm-rust/crates/core/src/audio_transcription/error.rs index f9ffb12d349..ab194173b67 100644 --- a/litellm-rust/crates/core/src/audio_transcription/error.rs +++ b/litellm-rust/crates/core/src/audio_transcription/error.rs @@ -24,3 +24,23 @@ pub enum Error { #[error(transparent)] Aws(#[from] litellm_auth_aws::Error), } + +impl From for Error { + fn from(error: litellm_providers::audio_transcription::Error) -> Self { + match error { + litellm_providers::audio_transcription::Error::InvalidType { expected, actual } => { + Self::InvalidType { expected, actual } + } + litellm_providers::audio_transcription::Error::MissingField(field) => { + Self::MissingField(field) + } + litellm_providers::audio_transcription::Error::InvalidRequest(message) => { + Self::InvalidRequest(message) + } + litellm_providers::audio_transcription::Error::InvalidResponse(message) => { + Self::InvalidResponse(message) + } + litellm_providers::audio_transcription::Error::Auth(error) => Self::Auth(error), + } + } +} diff --git a/litellm-rust/crates/core/src/audio_transcription/handler.rs b/litellm-rust/crates/core/src/audio_transcription/handler.rs index 4c48b6b5ede..ae547f10f15 100644 --- a/litellm-rust/crates/core/src/audio_transcription/handler.rs +++ b/litellm-rust/crates/core/src/audio_transcription/handler.rs @@ -47,8 +47,8 @@ async fn signed_headers( use std::collections::BTreeMap; use std::time::SystemTime; - use crate::llms::base_llm::audio_transcription::transformation::AudioTranscriptionAuth; use litellm_auth_aws::{aws_auth_config, resolve_credentials, sign_bedrock_post}; + use litellm_providers::base_llm::audio_transcription::transformation::AudioTranscriptionAuth; let AudioTranscriptionAuth::AwsSigV4 { region, .. } = &request.auth else { return Ok(request.upstream_headers.clone()); diff --git a/litellm-rust/crates/core/src/audio_transcription/mod.rs b/litellm-rust/crates/core/src/audio_transcription/mod.rs index fafc29a2d2a..b71e8d38b8a 100644 --- a/litellm-rust/crates/core/src/audio_transcription/mod.rs +++ b/litellm-rust/crates/core/src/audio_transcription/mod.rs @@ -3,7 +3,7 @@ pub use error::Error; mod client; mod handler; mod prepare; -pub mod types; +pub use litellm_providers::audio_transcription::types; pub use handler::execute_audio_transcription_provider_call; pub use prepare::prepare_audio_transcription_provider_call; diff --git a/litellm-rust/crates/core/src/audio_transcription/prepare.rs b/litellm-rust/crates/core/src/audio_transcription/prepare.rs index 26e705408e0..4dc3ffae191 100644 --- a/litellm-rust/crates/core/src/audio_transcription/prepare.rs +++ b/litellm-rust/crates/core/src/audio_transcription/prepare.rs @@ -4,10 +4,10 @@ use crate::http_utils::{has_header, string_headers}; use crate::litellm_core_utils::get_llm_provider_logic::{ CustomLlmProvider, get_custom_llm_provider, }; -use crate::llms::base_llm::audio_transcription::transformation::{ +use litellm_providers::base_llm::audio_transcription::transformation::{ AudioTranscriptionAuth, BaseAudioTranscriptionConfig, }; -use crate::llms::bedrock::audio_transcription::BEDROCK_AUDIO_TRANSCRIPTION_CONFIG; +use litellm_providers::bedrock::audio_transcription::BEDROCK_AUDIO_TRANSCRIPTION_CONFIG; fn provider_config(provider: &str) -> Option<&'static dyn BaseAudioTranscriptionConfig> { if provider == "bedrock" { diff --git a/litellm-rust/crates/core/src/chat_completions/common_utils.rs b/litellm-rust/crates/core/src/chat_completions/common_utils.rs index 8b966c7a173..63fc899e6f4 100644 --- a/litellm-rust/crates/core/src/chat_completions/common_utils.rs +++ b/litellm-rust/crates/core/src/chat_completions/common_utils.rs @@ -2,8 +2,8 @@ use serde_json::{Map, Value}; use super::Error; use crate::http_utils::string_headers as shared_string_headers; -use crate::llms::anthropic::chat::transformation::ANTHROPIC_CHAT_COMPLETIONS_CONFIG; -use crate::llms::base_llm::chat::transformation::BaseConfig; +use litellm_providers::anthropic::chat::transformation::ANTHROPIC_CHAT_COMPLETIONS_CONFIG; +use litellm_providers::base_llm::chat::transformation::BaseConfig; const HEADER_CONTEXT: &str = "chat completions"; @@ -11,7 +11,7 @@ pub(super) fn chat_completions_provider_config(provider: &str) -> Option<&'stati match provider { "anthropic" => Some(&ANTHROPIC_CHAT_COMPLETIONS_CONFIG), "bedrock" => Some( - &crate::llms::bedrock::chat::converse_transformation::BEDROCK_CHAT_COMPLETIONS_CONFIG, + &litellm_providers::bedrock::chat::converse_transformation::BEDROCK_CHAT_COMPLETIONS_CONFIG, ), _ => None, } diff --git a/litellm-rust/crates/core/src/chat_completions/error.rs b/litellm-rust/crates/core/src/chat_completions/error.rs index f9ffb12d349..95da97125d7 100644 --- a/litellm-rust/crates/core/src/chat_completions/error.rs +++ b/litellm-rust/crates/core/src/chat_completions/error.rs @@ -24,3 +24,19 @@ pub enum Error { #[error(transparent)] Aws(#[from] litellm_auth_aws::Error), } + +impl From for Error { + fn from(error: litellm_providers::chat::Error) -> Self { + match error { + litellm_providers::chat::Error::MissingField(field) => Self::MissingField(field), + litellm_providers::chat::Error::InvalidRequest(message) => { + Self::InvalidRequest(message) + } + litellm_providers::chat::Error::InvalidResponse(message) => { + Self::InvalidResponse(message) + } + litellm_providers::chat::Error::Unsupported(reason) => Self::Unsupported(reason), + litellm_providers::chat::Error::Auth(error) => Self::Auth(error), + } + } +} diff --git a/litellm-rust/crates/core/src/chat_completions/handler.rs b/litellm-rust/crates/core/src/chat_completions/handler.rs index 5090d481f6f..ac9f58cda22 100644 --- a/litellm-rust/crates/core/src/chat_completions/handler.rs +++ b/litellm-rust/crates/core/src/chat_completions/handler.rs @@ -8,7 +8,7 @@ use super::types::{ ResolvedChatCompletionsRequest, }; use crate::http_utils::{http_request, truncate_error_body}; -use crate::llms::base_llm::chat::transformation::ChatCompletionsAuth; +use litellm_providers::base_llm::chat::transformation::ChatCompletionsAuth; pub(super) async fn execute_chat_completions_provider_call( request: ResolvedChatCompletionsRequest<'_>, @@ -59,6 +59,7 @@ pub(super) async fn execute_chat_completions_provider_call( request .config .transform_response(&request.model, ProviderChatResponseData { body }) + .map_err(Error::from) .map_err(as_response_error) } diff --git a/litellm-rust/crates/core/src/chat_completions/mod.rs b/litellm-rust/crates/core/src/chat_completions/mod.rs index 5f7448bf73a..2215e1d9c5b 100644 --- a/litellm-rust/crates/core/src/chat_completions/mod.rs +++ b/litellm-rust/crates/core/src/chat_completions/mod.rs @@ -10,12 +10,11 @@ mod error; pub use error::Error; mod client; mod common_utils; -pub mod conversation; +pub use litellm_providers::chat::{conversation, response_utils}; pub(crate) mod handler; mod prepare; -pub mod response_utils; pub mod streaming; -pub mod types; +pub use litellm_providers::chat::types; use handler::execute_chat_completions_provider_call; use prepare::{parse_messages, resolve_provider_config, resolve_request}; diff --git a/litellm-rust/crates/core/src/chat_completions/prepare.rs b/litellm-rust/crates/core/src/chat_completions/prepare.rs index 983fbdf4f1d..3f3f97d6191 100644 --- a/litellm-rust/crates/core/src/chat_completions/prepare.rs +++ b/litellm-rust/crates/core/src/chat_completions/prepare.rs @@ -10,7 +10,7 @@ use crate::http_utils::has_header; use crate::litellm_core_utils::get_llm_provider_logic::{ CustomLlmProvider, get_custom_llm_provider, }; -use crate::llms::base_llm::chat::transformation::{BaseConfig, ChatCompletionsAuth}; +use litellm_providers::base_llm::chat::transformation::{BaseConfig, ChatCompletionsAuth}; pub(super) fn resolve_provider_config<'a>( model: &'a str, diff --git a/litellm-rust/crates/core/src/chat_completions/tests.rs b/litellm-rust/crates/core/src/chat_completions/tests.rs index 86ac6c6ca35..40298cf5c2e 100644 --- a/litellm-rust/crates/core/src/chat_completions/tests.rs +++ b/litellm-rust/crates/core/src/chat_completions/tests.rs @@ -3,7 +3,7 @@ use serde_json::{Map, Value, json}; use super::Error; use super::prepare::{prepare_provider_request, resolve_request}; use super::types::{ChatCompletionsRequest, ProviderChatCompletionsRequest}; -use crate::llms::base_llm::chat::transformation::ChatCompletionsAuth; +use litellm_providers::base_llm::chat::transformation::ChatCompletionsAuth; fn prepare_chat_completions_call( request: ChatCompletionsRequest<'_>, diff --git a/litellm-rust/crates/core/src/litellm_core_utils/get_llm_provider_logic.rs b/litellm-rust/crates/core/src/litellm_core_utils/get_llm_provider_logic.rs index 6333eedebfc..5958e8ac613 100644 --- a/litellm-rust/crates/core/src/litellm_core_utils/get_llm_provider_logic.rs +++ b/litellm-rust/crates/core/src/litellm_core_utils/get_llm_provider_logic.rs @@ -1,36 +1,4 @@ -#[derive(Debug, Clone, Copy, PartialEq, Eq)] -pub struct CustomLlmProvider<'a> { - pub model: &'a str, - pub custom_llm_provider: &'a str, -} - -pub fn get_custom_llm_provider<'a>( - model: &'a str, - custom_llm_provider: Option<&'a str>, -) -> Option> { - if let Some(custom_llm_provider) = custom_llm_provider.filter(|provider| !provider.is_empty()) { - return Some(CustomLlmProvider { - model: strip_custom_llm_provider_prefix(model, custom_llm_provider), - custom_llm_provider, - }); - } - - let (custom_llm_provider, model) = model.split_once('/')?; - if custom_llm_provider.is_empty() || model.is_empty() { - return None; - } - Some(CustomLlmProvider { - model, - custom_llm_provider, - }) -} - -fn strip_custom_llm_provider_prefix<'a>(model: &'a str, custom_llm_provider: &str) -> &'a str { - model - .strip_prefix(custom_llm_provider) - .and_then(|model| model.strip_prefix('/')) - .unwrap_or(model) -} +pub use litellm_providers::provider_resolution::{CustomLlmProvider, get_custom_llm_provider}; #[cfg(test)] mod tests { diff --git a/litellm-rust/crates/core/src/llms/anthropic/chat/mod.rs b/litellm-rust/crates/core/src/llms/anthropic/chat/mod.rs index fa7df180f50..7bf4fc46291 100644 --- a/litellm-rust/crates/core/src/llms/anthropic/chat/mod.rs +++ b/litellm-rust/crates/core/src/llms/anthropic/chat/mod.rs @@ -1,2 +1 @@ pub mod streaming; -pub mod transformation; diff --git a/litellm-rust/crates/core/src/llms/anthropic/chat/streaming.rs b/litellm-rust/crates/core/src/llms/anthropic/chat/streaming.rs index ec44f2c5808..427c57633d3 100644 --- a/litellm-rust/crates/core/src/llms/anthropic/chat/streaming.rs +++ b/litellm-rust/crates/core/src/llms/anthropic/chat/streaming.rs @@ -2,16 +2,16 @@ use std::collections::HashMap; use serde_json::Value; +use super::super::experimental_pass_through::messages::streaming::{ + AnthropicContentBlock, AnthropicContentBlockDelta, AnthropicMessagesStreamEvent, + AnthropicStreamUsage, +}; use crate::chat_completions::Error; use crate::chat_completions::streaming::StreamTransformer; use crate::chat_completions::types::{ ChatCompletionChunk, ChatCompletionThinkingBlock, ChatCompletionToolCallChunk, ChatCompletionsUsage, }; -use crate::llms::anthropic::experimental_pass_through::messages::streaming::{ - AnthropicContentBlock, AnthropicContentBlockDelta, AnthropicMessagesStreamEvent, - AnthropicStreamUsage, -}; #[derive(Clone, Copy, Debug, Eq, PartialEq)] pub enum AnthropicJsonChunkType { diff --git a/litellm-rust/crates/core/src/llms/anthropic/experimental_pass_through/messages/batches.rs b/litellm-rust/crates/core/src/llms/anthropic/experimental_pass_through/messages/batches.rs index 762e699acba..16b4e2a59ad 100644 --- a/litellm-rust/crates/core/src/llms/anthropic/experimental_pass_through/messages/batches.rs +++ b/litellm-rust/crates/core/src/llms/anthropic/experimental_pass_through/messages/batches.rs @@ -3,9 +3,9 @@ use serde_json::Value; use time::OffsetDateTime; use url::Url; -use crate::llms::anthropic::experimental_pass_through::messages::transformation::resolve_anthropic_api_base; use crate::messages::Error; use crate::messages::types::AnthropicMessagesResponse; +use litellm_providers::anthropic::experimental_pass_through::messages::transformation::resolve_anthropic_api_base; const BATCHES_PATH_SUFFIX: &str = "/v1/messages/batches"; diff --git a/litellm-rust/crates/core/src/llms/anthropic/experimental_pass_through/messages/mod.rs b/litellm-rust/crates/core/src/llms/anthropic/experimental_pass_through/messages/mod.rs index 3b1da7dc069..42d4fcdde0f 100644 --- a/litellm-rust/crates/core/src/llms/anthropic/experimental_pass_through/messages/mod.rs +++ b/litellm-rust/crates/core/src/llms/anthropic/experimental_pass_through/messages/mod.rs @@ -1,4 +1,3 @@ pub mod batches; pub mod count_tokens; pub mod streaming; -pub mod transformation; diff --git a/litellm-rust/crates/core/src/llms/azure_ai/mod.rs b/litellm-rust/crates/core/src/llms/azure_ai/mod.rs index 8a52bda45be..079e0c41eae 100644 --- a/litellm-rust/crates/core/src/llms/azure_ai/mod.rs +++ b/litellm-rust/crates/core/src/llms/azure_ai/mod.rs @@ -1,2 +1 @@ -pub mod anthropic; pub(crate) mod ocr; diff --git a/litellm-rust/crates/core/src/llms/base_llm/mod.rs b/litellm-rust/crates/core/src/llms/base_llm/mod.rs index 5cd48a21fb6..079e0c41eae 100644 --- a/litellm-rust/crates/core/src/llms/base_llm/mod.rs +++ b/litellm-rust/crates/core/src/llms/base_llm/mod.rs @@ -1,4 +1 @@ -pub mod anthropic_messages; -pub mod audio_transcription; -pub mod chat; pub(crate) mod ocr; diff --git a/litellm-rust/crates/core/src/llms/mod.rs b/litellm-rust/crates/core/src/llms/mod.rs index 635d381561c..4b93a5f971c 100644 --- a/litellm-rust/crates/core/src/llms/mod.rs +++ b/litellm-rust/crates/core/src/llms/mod.rs @@ -1,7 +1,6 @@ pub mod anthropic; pub mod azure_ai; pub mod base_llm; -pub mod bedrock; pub(crate) mod cohere; pub(crate) mod mistral; pub mod openai; diff --git a/litellm-rust/crates/core/src/messages/common_utils.rs b/litellm-rust/crates/core/src/messages/common_utils.rs index a0a120c34a9..c58e9122cad 100644 --- a/litellm-rust/crates/core/src/messages/common_utils.rs +++ b/litellm-rust/crates/core/src/messages/common_utils.rs @@ -3,9 +3,9 @@ use serde_json::{Map, Value}; use super::Error; use crate::http_utils::string_headers as shared_string_headers; pub(super) use crate::http_utils::{has_bearer_auth, has_header, truncate_error_body}; -use crate::llms::anthropic::experimental_pass_through::messages::transformation::ANTHROPIC_MESSAGES_CONFIG; -use crate::llms::azure_ai::anthropic::messages_transformation::AZURE_ANTHROPIC_MESSAGES_CONFIG; -use crate::llms::base_llm::anthropic_messages::transformation::BaseAnthropicMessagesConfig; +use litellm_providers::anthropic::experimental_pass_through::messages::transformation::ANTHROPIC_MESSAGES_CONFIG; +use litellm_providers::azure_ai::anthropic::messages_transformation::AZURE_ANTHROPIC_MESSAGES_CONFIG; +use litellm_providers::base_llm::anthropic_messages::transformation::BaseAnthropicMessagesConfig; const HEADER_CONTEXT: &str = "messages"; diff --git a/litellm-rust/crates/core/src/messages/error.rs b/litellm-rust/crates/core/src/messages/error.rs index f5e86c4850e..cdb4de4645f 100644 --- a/litellm-rust/crates/core/src/messages/error.rs +++ b/litellm-rust/crates/core/src/messages/error.rs @@ -28,6 +28,22 @@ pub enum Error { InvalidBedrockBase64(String), } +impl From for Error { + fn from(error: litellm_providers::messages::Error) -> Self { + match error { + litellm_providers::messages::Error::MissingField(field) => Self::MissingField(field), + litellm_providers::messages::Error::InvalidRequest(message) => { + Self::InvalidRequest(message) + } + litellm_providers::messages::Error::InvalidResponse(message) => { + Self::InvalidResponse(message) + } + litellm_providers::messages::Error::Unsupported(reason) => Self::Unsupported(reason), + litellm_providers::messages::Error::Auth(error) => Self::Auth(error), + } + } +} + impl Error { pub fn is_request(&self) -> bool { match self { diff --git a/litellm-rust/crates/core/src/messages/handler.rs b/litellm-rust/crates/core/src/messages/handler.rs index 36d8d2f0157..e241bc56c1e 100644 --- a/litellm-rust/crates/core/src/messages/handler.rs +++ b/litellm-rust/crates/core/src/messages/handler.rs @@ -40,6 +40,7 @@ pub(super) async fn execute_messages_provider_call( request .config .transform_anthropic_messages_response(&request.model, response) + .map_err(Error::from) } pub(super) async fn execute_messages_provider_stream( diff --git a/litellm-rust/crates/core/src/messages/mod.rs b/litellm-rust/crates/core/src/messages/mod.rs index 8f6fffcaf7f..5149c52478d 100644 --- a/litellm-rust/crates/core/src/messages/mod.rs +++ b/litellm-rust/crates/core/src/messages/mod.rs @@ -13,7 +13,7 @@ mod client; mod common_utils; mod handler; mod prepare; -pub mod types; +pub use litellm_providers::messages::types; use handler::{execute_messages_provider_call, execute_messages_provider_stream}; use types::{AnthropicMessagesResponse, MessagesRequest}; diff --git a/litellm-rust/crates/core/src/messages/prepare.rs b/litellm-rust/crates/core/src/messages/prepare.rs index a3c93746d3e..f3735ff1700 100644 --- a/litellm-rust/crates/core/src/messages/prepare.rs +++ b/litellm-rust/crates/core/src/messages/prepare.rs @@ -6,7 +6,7 @@ use super::types::{MessagesRequest, ProviderMessagesRequest}; use crate::litellm_core_utils::get_llm_provider_logic::{ CustomLlmProvider, get_custom_llm_provider, }; -use crate::llms::base_llm::anthropic_messages::transformation::{ +use litellm_providers::base_llm::anthropic_messages::transformation::{ BaseAnthropicMessagesConfig, MessagesAuthStrategy, }; diff --git a/litellm-rust/crates/providers/Cargo.toml b/litellm-rust/crates/providers/Cargo.toml new file mode 100644 index 00000000000..e1c8f2c50d4 --- /dev/null +++ b/litellm-rust/crates/providers/Cargo.toml @@ -0,0 +1,16 @@ +[package] +name = "litellm-providers" +version = "0.1.0" +edition.workspace = true +license.workspace = true +repository.workspace = true + +[dependencies] +litellm-auth.workspace = true +litellm-auth-aws.workspace = true +serde.workspace = true +serde_json.workspace = true +thiserror.workspace = true + +[dev-dependencies] +rstest.workspace = true diff --git a/litellm-rust/crates/core/src/llms/base_llm/anthropic_messages/mod.rs b/litellm-rust/crates/providers/src/anthropic/chat/mod.rs similarity index 100% rename from litellm-rust/crates/core/src/llms/base_llm/anthropic_messages/mod.rs rename to litellm-rust/crates/providers/src/anthropic/chat/mod.rs diff --git a/litellm-rust/crates/core/src/llms/anthropic/chat/tests.rs b/litellm-rust/crates/providers/src/anthropic/chat/tests.rs similarity index 99% rename from litellm-rust/crates/core/src/llms/anthropic/chat/tests.rs rename to litellm-rust/crates/providers/src/anthropic/chat/tests.rs index 25c2f5e49f4..18b6efb13fd 100644 --- a/litellm-rust/crates/core/src/llms/anthropic/chat/tests.rs +++ b/litellm-rust/crates/providers/src/anthropic/chat/tests.rs @@ -1,7 +1,7 @@ use serde_json::json; use super::*; -use crate::chat_completions::Error; +use crate::chat::Error; fn messages(value: Value) -> Vec { serde_json::from_value(value).expect("valid messages") diff --git a/litellm-rust/crates/core/src/llms/anthropic/chat/transformation.rs b/litellm-rust/crates/providers/src/anthropic/chat/transformation.rs similarity index 94% rename from litellm-rust/crates/core/src/llms/anthropic/chat/transformation.rs rename to litellm-rust/crates/providers/src/anthropic/chat/transformation.rs index fc48ef6d74f..5288eebbb2f 100644 --- a/litellm-rust/crates/core/src/llms/anthropic/chat/transformation.rs +++ b/litellm-rust/crates/providers/src/anthropic/chat/transformation.rs @@ -1,19 +1,19 @@ use serde_json::{Map, Value, json}; -use crate::chat_completions::Error; -use crate::chat_completions::conversation::{Conversation, build_conversation}; -use crate::chat_completions::response_utils::{finish_reason_for, unix_now, usage_from_parts}; -use crate::chat_completions::types::{ - ChatCompletionsChoice, ChatCompletionsChoiceMessage, ChatCompletionsResponse, ChatMessage, - ProviderChatRequestData, ProviderChatResponseData, -}; -use crate::constants::ANTHROPIC_OAUTH_TOKEN_PREFIX; -use crate::llms::anthropic::experimental_pass_through::messages::transformation::{ +use crate::anthropic::ANTHROPIC_OAUTH_TOKEN_PREFIX; +use crate::anthropic::experimental_pass_through::messages::transformation::{ complete_anthropic_url, resolve_anthropic_api_key, }; -use crate::llms::base_llm::chat::transformation::{ +use crate::base_llm::chat::transformation::{ BaseConfig, ChatCompletionsAuth, Unsupported, unsupported_message, unsupported_param, }; +use crate::chat::Error; +use crate::chat::conversation::{Conversation, build_conversation}; +use crate::chat::response_utils::{finish_reason_for, unix_now, usage_from_parts}; +use crate::chat::types::{ + ChatCompletionsChoice, ChatCompletionsChoiceMessage, ChatCompletionsResponse, ChatMessage, + ProviderChatRequestData, ProviderChatResponseData, +}; /// Anthropic parameter names, post `map_openai_params`, that the Rust path can /// place verbatim in the Messages body. diff --git a/litellm-rust/crates/core/src/llms/base_llm/audio_transcription/mod.rs b/litellm-rust/crates/providers/src/anthropic/experimental_pass_through/messages/mod.rs similarity index 100% rename from litellm-rust/crates/core/src/llms/base_llm/audio_transcription/mod.rs rename to litellm-rust/crates/providers/src/anthropic/experimental_pass_through/messages/mod.rs diff --git a/litellm-rust/crates/core/src/llms/anthropic/experimental_pass_through/messages/transformation.rs b/litellm-rust/crates/providers/src/anthropic/experimental_pass_through/messages/transformation.rs similarity index 97% rename from litellm-rust/crates/core/src/llms/anthropic/experimental_pass_through/messages/transformation.rs rename to litellm-rust/crates/providers/src/anthropic/experimental_pass_through/messages/transformation.rs index 6a1a1613ffc..beabe440269 100644 --- a/litellm-rust/crates/core/src/llms/anthropic/experimental_pass_through/messages/transformation.rs +++ b/litellm-rust/crates/providers/src/anthropic/experimental_pass_through/messages/transformation.rs @@ -1,4 +1,4 @@ -use crate::llms::base_llm::anthropic_messages::transformation::BaseAnthropicMessagesConfig; +use crate::base_llm::anthropic_messages::transformation::BaseAnthropicMessagesConfig; use crate::messages::Error; const ANTHROPIC_API_KEY_ENV: &str = "ANTHROPIC_API_KEY"; diff --git a/litellm-rust/crates/providers/src/anthropic/experimental_pass_through/mod.rs b/litellm-rust/crates/providers/src/anthropic/experimental_pass_through/mod.rs new file mode 100644 index 00000000000..ba63992f3cb --- /dev/null +++ b/litellm-rust/crates/providers/src/anthropic/experimental_pass_through/mod.rs @@ -0,0 +1 @@ +pub mod messages; diff --git a/litellm-rust/crates/providers/src/anthropic/mod.rs b/litellm-rust/crates/providers/src/anthropic/mod.rs new file mode 100644 index 00000000000..38a59aa6e0d --- /dev/null +++ b/litellm-rust/crates/providers/src/anthropic/mod.rs @@ -0,0 +1,4 @@ +pub mod chat; +pub mod experimental_pass_through; + +pub const ANTHROPIC_OAUTH_TOKEN_PREFIX: &str = "sk-ant-oat"; diff --git a/litellm-rust/crates/providers/src/audio_transcription/mod.rs b/litellm-rust/crates/providers/src/audio_transcription/mod.rs new file mode 100644 index 00000000000..278b049e8f9 --- /dev/null +++ b/litellm-rust/crates/providers/src/audio_transcription/mod.rs @@ -0,0 +1,31 @@ +use thiserror::Error; + +#[derive(Clone, Debug, PartialEq, Eq, Error)] +pub enum Error { + #[error("expected {expected}, got {actual}")] + InvalidType { + expected: &'static str, + actual: &'static str, + }, + #[error("missing required field: {0}")] + MissingField(&'static str), + #[error("invalid request: {0}")] + InvalidRequest(String), + #[error("invalid response: {0}")] + InvalidResponse(String), + #[error(transparent)] + Auth(#[from] litellm_auth::Error), +} + +pub fn json_type_name(value: &serde_json::Value) -> &'static str { + match value { + serde_json::Value::Null => "null", + serde_json::Value::Bool(_) => "boolean", + serde_json::Value::Number(_) => "number", + serde_json::Value::String(_) => "string", + serde_json::Value::Array(_) => "array", + serde_json::Value::Object(_) => "object", + } +} + +pub mod types; diff --git a/litellm-rust/crates/core/src/audio_transcription/types.rs b/litellm-rust/crates/providers/src/audio_transcription/types.rs similarity index 74% rename from litellm-rust/crates/core/src/audio_transcription/types.rs rename to litellm-rust/crates/providers/src/audio_transcription/types.rs index 1ec1f224f6b..d17d5067de5 100644 --- a/litellm-rust/crates/core/src/audio_transcription/types.rs +++ b/litellm-rust/crates/providers/src/audio_transcription/types.rs @@ -3,7 +3,7 @@ use std::time::Duration; use serde::{Deserialize, Serialize}; use serde_json::{Map, Value}; -use crate::llms::base_llm::audio_transcription::transformation::{ +use crate::base_llm::audio_transcription::transformation::{ AudioTranscriptionAuth, BaseAudioTranscriptionConfig, }; @@ -20,15 +20,15 @@ pub struct AudioTranscriptionRequest<'a> { #[derive(Clone)] pub struct ProviderAudioTranscriptionRequest { - pub(super) model: String, - pub(super) custom_llm_provider: String, - pub(super) config: &'static dyn BaseAudioTranscriptionConfig, - pub(super) url: String, - pub(super) body: Value, - pub(super) upstream_headers: Vec<(String, String)>, - pub(super) auth: AudioTranscriptionAuth, - pub(super) optional_params: Map, - pub(super) timeout: Option, + pub model: String, + pub custom_llm_provider: String, + pub config: &'static dyn BaseAudioTranscriptionConfig, + pub url: String, + pub body: Value, + pub upstream_headers: Vec<(String, String)>, + pub auth: AudioTranscriptionAuth, + pub optional_params: Map, + pub timeout: Option, } impl ProviderAudioTranscriptionRequest { diff --git a/litellm-rust/crates/core/src/llms/azure_ai/anthropic/messages_transformation.rs b/litellm-rust/crates/providers/src/azure_ai/anthropic/messages_transformation.rs similarity index 99% rename from litellm-rust/crates/core/src/llms/azure_ai/anthropic/messages_transformation.rs rename to litellm-rust/crates/providers/src/azure_ai/anthropic/messages_transformation.rs index feaee0375c4..a79b9038144 100644 --- a/litellm-rust/crates/core/src/llms/azure_ai/anthropic/messages_transformation.rs +++ b/litellm-rust/crates/providers/src/azure_ai/anthropic/messages_transformation.rs @@ -1,9 +1,9 @@ use serde_json::{Map, Value}; -use crate::llms::anthropic::experimental_pass_through::messages::transformation::{ +use crate::anthropic::experimental_pass_through::messages::transformation::{ ANTHROPIC_MESSAGES_CONFIG, AnthropicMessagesConfig, non_empty, }; -use crate::llms::base_llm::anthropic_messages::transformation::{ +use crate::base_llm::anthropic_messages::transformation::{ BaseAnthropicMessagesConfig, MessagesAuthStrategy, }; use crate::messages::Error; diff --git a/litellm-rust/crates/core/src/llms/azure_ai/anthropic/mod.rs b/litellm-rust/crates/providers/src/azure_ai/anthropic/mod.rs similarity index 100% rename from litellm-rust/crates/core/src/llms/azure_ai/anthropic/mod.rs rename to litellm-rust/crates/providers/src/azure_ai/anthropic/mod.rs diff --git a/litellm-rust/crates/providers/src/azure_ai/mod.rs b/litellm-rust/crates/providers/src/azure_ai/mod.rs new file mode 100644 index 00000000000..e529997219e --- /dev/null +++ b/litellm-rust/crates/providers/src/azure_ai/mod.rs @@ -0,0 +1 @@ +pub mod anthropic; diff --git a/litellm-rust/crates/core/src/llms/base_llm/chat/mod.rs b/litellm-rust/crates/providers/src/base_llm/anthropic_messages/mod.rs similarity index 100% rename from litellm-rust/crates/core/src/llms/base_llm/chat/mod.rs rename to litellm-rust/crates/providers/src/base_llm/anthropic_messages/mod.rs diff --git a/litellm-rust/crates/core/src/llms/base_llm/anthropic_messages/transformation.rs b/litellm-rust/crates/providers/src/base_llm/anthropic_messages/transformation.rs similarity index 100% rename from litellm-rust/crates/core/src/llms/base_llm/anthropic_messages/transformation.rs rename to litellm-rust/crates/providers/src/base_llm/anthropic_messages/transformation.rs diff --git a/litellm-rust/crates/providers/src/base_llm/audio_transcription/mod.rs b/litellm-rust/crates/providers/src/base_llm/audio_transcription/mod.rs new file mode 100644 index 00000000000..f239b6921fa --- /dev/null +++ b/litellm-rust/crates/providers/src/base_llm/audio_transcription/mod.rs @@ -0,0 +1 @@ +pub mod transformation; diff --git a/litellm-rust/crates/core/src/llms/base_llm/audio_transcription/transformation.rs b/litellm-rust/crates/providers/src/base_llm/audio_transcription/transformation.rs similarity index 100% rename from litellm-rust/crates/core/src/llms/base_llm/audio_transcription/transformation.rs rename to litellm-rust/crates/providers/src/base_llm/audio_transcription/transformation.rs diff --git a/litellm-rust/crates/providers/src/base_llm/chat/mod.rs b/litellm-rust/crates/providers/src/base_llm/chat/mod.rs new file mode 100644 index 00000000000..f239b6921fa --- /dev/null +++ b/litellm-rust/crates/providers/src/base_llm/chat/mod.rs @@ -0,0 +1 @@ +pub mod transformation; diff --git a/litellm-rust/crates/core/src/llms/base_llm/chat/transformation.rs b/litellm-rust/crates/providers/src/base_llm/chat/transformation.rs similarity index 98% rename from litellm-rust/crates/core/src/llms/base_llm/chat/transformation.rs rename to litellm-rust/crates/providers/src/base_llm/chat/transformation.rs index cb340db7326..5d81dc1a85e 100644 --- a/litellm-rust/crates/core/src/llms/base_llm/chat/transformation.rs +++ b/litellm-rust/crates/providers/src/base_llm/chat/transformation.rs @@ -1,7 +1,7 @@ use serde_json::{Map, Value}; -use crate::chat_completions::Error; -use crate::chat_completions::types::{ +use crate::chat::Error; +use crate::chat::types::{ ChatCompletionsResponse, ChatMessage, ChatMessageContent, ProviderChatRequestData, ProviderChatResponseData, }; diff --git a/litellm-rust/crates/providers/src/base_llm/mod.rs b/litellm-rust/crates/providers/src/base_llm/mod.rs new file mode 100644 index 00000000000..b7a1f696440 --- /dev/null +++ b/litellm-rust/crates/providers/src/base_llm/mod.rs @@ -0,0 +1,3 @@ +pub mod anthropic_messages; +pub mod audio_transcription; +pub mod chat; diff --git a/litellm-rust/crates/core/src/llms/bedrock/audio_transcription/mod.rs b/litellm-rust/crates/providers/src/bedrock/audio_transcription/mod.rs similarity index 98% rename from litellm-rust/crates/core/src/llms/bedrock/audio_transcription/mod.rs rename to litellm-rust/crates/providers/src/bedrock/audio_transcription/mod.rs index 49397e00901..7da2aa42a51 100644 --- a/litellm-rust/crates/core/src/llms/bedrock/audio_transcription/mod.rs +++ b/litellm-rust/crates/providers/src/bedrock/audio_transcription/mod.rs @@ -1,11 +1,11 @@ use serde_json::{Map, Value, json}; use crate::audio_transcription::Error; +use crate::audio_transcription::json_type_name; use crate::audio_transcription::types::{ AudioTranscriptionRequestData, AudioTranscriptionResponseData, }; -use crate::http_utils::json_type_name; -use crate::llms::base_llm::audio_transcription::transformation::{ +use crate::base_llm::audio_transcription::transformation::{ AudioTranscriptionAuth, BaseAudioTranscriptionConfig, }; use litellm_auth_aws::constants::{BEDROCK_RUNTIME_ENDPOINT_TEMPLATE, BEDROCK_SERVICE}; diff --git a/litellm-rust/crates/core/src/llms/bedrock/chat/converse_transformation.rs b/litellm-rust/crates/providers/src/bedrock/chat/converse_transformation.rs similarity index 97% rename from litellm-rust/crates/core/src/llms/bedrock/chat/converse_transformation.rs rename to litellm-rust/crates/providers/src/bedrock/chat/converse_transformation.rs index 525bb6d7abc..85ba3be9b07 100644 --- a/litellm-rust/crates/core/src/llms/bedrock/chat/converse_transformation.rs +++ b/litellm-rust/crates/providers/src/bedrock/chat/converse_transformation.rs @@ -1,16 +1,16 @@ use serde_json::{Map, Value, json}; -use crate::chat_completions::Error; -use crate::chat_completions::conversation::{Conversation, TurnRole, build_conversation}; -use crate::chat_completions::response_utils::{finish_reason_for, unix_now, usage_from_parts}; -use crate::chat_completions::types::{ +use crate::base_llm::chat::transformation::{ + BaseConfig, ChatCompletionsAuth, Unsupported, unsupported_message, unsupported_param, +}; +use crate::chat::Error; +use crate::chat::conversation::{Conversation, TurnRole, build_conversation}; +use crate::chat::response_utils::{finish_reason_for, unix_now, usage_from_parts}; +use crate::chat::types::{ ChatCompletionsChoice, ChatCompletionsChoiceMessage, ChatCompletionsResponse, ChatCompletionsUsage, ChatMessage, ChatMessageContent, ProviderChatRequestData, ProviderChatResponseData, }; -use crate::llms::base_llm::chat::transformation::{ - BaseConfig, ChatCompletionsAuth, Unsupported, unsupported_message, unsupported_param, -}; use litellm_auth_aws::constants::{AWS_BEARER_TOKEN_BEDROCK, BEDROCK_RUNTIME_ENDPOINT_TEMPLATE}; use litellm_auth_aws::{bedrock_model_id_and_region, resolve_bedrock_region}; diff --git a/litellm-rust/crates/core/src/llms/bedrock/chat/mod.rs b/litellm-rust/crates/providers/src/bedrock/chat/mod.rs similarity index 100% rename from litellm-rust/crates/core/src/llms/bedrock/chat/mod.rs rename to litellm-rust/crates/providers/src/bedrock/chat/mod.rs diff --git a/litellm-rust/crates/core/src/llms/bedrock/chat/tests.rs b/litellm-rust/crates/providers/src/bedrock/chat/tests.rs similarity index 99% rename from litellm-rust/crates/core/src/llms/bedrock/chat/tests.rs rename to litellm-rust/crates/providers/src/bedrock/chat/tests.rs index ed34a46c431..cfa0c902096 100644 --- a/litellm-rust/crates/core/src/llms/bedrock/chat/tests.rs +++ b/litellm-rust/crates/providers/src/bedrock/chat/tests.rs @@ -1,7 +1,7 @@ use serde_json::json; use super::*; -use crate::chat_completions::Error; +use crate::chat::Error; fn messages(value: Value) -> Vec { serde_json::from_value(value).expect("valid messages") diff --git a/litellm-rust/crates/core/src/llms/bedrock/mod.rs b/litellm-rust/crates/providers/src/bedrock/mod.rs similarity index 100% rename from litellm-rust/crates/core/src/llms/bedrock/mod.rs rename to litellm-rust/crates/providers/src/bedrock/mod.rs diff --git a/litellm-rust/crates/core/src/chat_completions/conversation.rs b/litellm-rust/crates/providers/src/chat/conversation.rs similarity index 99% rename from litellm-rust/crates/core/src/chat_completions/conversation.rs rename to litellm-rust/crates/providers/src/chat/conversation.rs index 1f1984ed8be..587b7ea2a16 100644 --- a/litellm-rust/crates/core/src/chat_completions/conversation.rs +++ b/litellm-rust/crates/providers/src/chat/conversation.rs @@ -11,7 +11,7 @@ //! accepts; anything richer is declined upstream by the capability gate. use super::types::{ChatMessage, ChatMessageContent}; -use crate::constants::EMPTY_TEXT_PLACEHOLDER; +use crate::chat::EMPTY_TEXT_PLACEHOLDER; #[derive(Clone, Copy, Debug, PartialEq, Eq)] pub enum TurnRole { diff --git a/litellm-rust/crates/providers/src/chat/mod.rs b/litellm-rust/crates/providers/src/chat/mod.rs new file mode 100644 index 00000000000..93892657c75 --- /dev/null +++ b/litellm-rust/crates/providers/src/chat/mod.rs @@ -0,0 +1,21 @@ +use thiserror::Error; + +pub const EMPTY_TEXT_PLACEHOLDER: &str = " "; + +#[derive(Clone, Debug, PartialEq, Eq, Error)] +pub enum Error { + #[error("missing required field: {0}")] + MissingField(&'static str), + #[error("invalid request: {0}")] + InvalidRequest(String), + #[error("invalid response: {0}")] + InvalidResponse(String), + #[error("unsupported: {0}")] + Unsupported(&'static str), + #[error(transparent)] + Auth(#[from] litellm_auth::Error), +} + +pub mod conversation; +pub mod response_utils; +pub mod types; diff --git a/litellm-rust/crates/core/src/chat_completions/response_utils.rs b/litellm-rust/crates/providers/src/chat/response_utils.rs similarity index 100% rename from litellm-rust/crates/core/src/chat_completions/response_utils.rs rename to litellm-rust/crates/providers/src/chat/response_utils.rs diff --git a/litellm-rust/crates/core/src/chat_completions/types.rs b/litellm-rust/crates/providers/src/chat/types.rs similarity index 88% rename from litellm-rust/crates/core/src/chat_completions/types.rs rename to litellm-rust/crates/providers/src/chat/types.rs index 48bb0b12966..d61892624cf 100644 --- a/litellm-rust/crates/core/src/chat_completions/types.rs +++ b/litellm-rust/crates/providers/src/chat/types.rs @@ -3,7 +3,7 @@ use std::time::Duration; use serde::{Deserialize, Serialize}; use serde_json::{Map, Value}; -use crate::llms::base_llm::chat::transformation::{BaseConfig, ChatCompletionsAuth}; +use crate::base_llm::chat::transformation::{BaseConfig, ChatCompletionsAuth}; /// A `/chat/completions` call as it crosses into the core. /// @@ -22,26 +22,26 @@ pub struct ChatCompletionsRequest<'a> { pub timeout: Option, } -pub(super) struct ResolvedChatCompletionsRequest<'a> { - pub(super) model: String, - pub(super) config: &'static dyn BaseConfig, - pub(super) messages: Vec, - pub(super) optional_params: Map, - pub(super) api_key: Option<&'a str>, - pub(super) api_base: Option<&'a str>, - pub(super) extra_headers: Option>, - pub(super) timeout: Option, +pub struct ResolvedChatCompletionsRequest<'a> { + pub model: String, + pub config: &'static dyn BaseConfig, + pub messages: Vec, + pub optional_params: Map, + pub api_key: Option<&'a str>, + pub api_base: Option<&'a str>, + pub extra_headers: Option>, + pub timeout: Option, } -pub(super) struct ProviderChatCompletionsRequest { - pub(super) model: String, - pub(super) config: &'static dyn BaseConfig, - pub(super) url: String, - pub(super) body: Value, - pub(super) upstream_headers: Vec<(String, String)>, - pub(super) auth: ChatCompletionsAuth, - pub(super) optional_params: Map, - pub(super) timeout: Option, +pub struct ProviderChatCompletionsRequest { + pub model: String, + pub config: &'static dyn BaseConfig, + pub url: String, + pub body: Value, + pub upstream_headers: Vec<(String, String)>, + pub auth: ChatCompletionsAuth, + pub optional_params: Map, + pub timeout: Option, } /// The provider-shaped request body a config produces. Named rather than a bare diff --git a/litellm-rust/crates/providers/src/lib.rs b/litellm-rust/crates/providers/src/lib.rs new file mode 100644 index 00000000000..5d72ffffb2b --- /dev/null +++ b/litellm-rust/crates/providers/src/lib.rs @@ -0,0 +1,8 @@ +pub mod anthropic; +pub mod audio_transcription; +pub mod azure_ai; +pub mod base_llm; +pub mod bedrock; +pub mod chat; +pub mod messages; +pub mod provider_resolution; diff --git a/litellm-rust/crates/providers/src/messages/mod.rs b/litellm-rust/crates/providers/src/messages/mod.rs new file mode 100644 index 00000000000..07232b36b51 --- /dev/null +++ b/litellm-rust/crates/providers/src/messages/mod.rs @@ -0,0 +1,17 @@ +use thiserror::Error; + +#[derive(Clone, Debug, PartialEq, Eq, Error)] +pub enum Error { + #[error("missing required field: {0}")] + MissingField(&'static str), + #[error("invalid request: {0}")] + InvalidRequest(String), + #[error("invalid response: {0}")] + InvalidResponse(String), + #[error("unsupported: {0}")] + Unsupported(&'static str), + #[error(transparent)] + Auth(#[from] litellm_auth::Error), +} + +pub mod types; diff --git a/litellm-rust/crates/core/src/messages/types.rs b/litellm-rust/crates/providers/src/messages/types.rs similarity index 90% rename from litellm-rust/crates/core/src/messages/types.rs rename to litellm-rust/crates/providers/src/messages/types.rs index 32cf4b29faf..ba274ab9651 100644 --- a/litellm-rust/crates/core/src/messages/types.rs +++ b/litellm-rust/crates/providers/src/messages/types.rs @@ -3,7 +3,7 @@ use std::time::Duration; use serde::{Deserialize, Serialize}; use serde_json::{Map, Value}; -use crate::llms::base_llm::anthropic_messages::transformation::BaseAnthropicMessagesConfig; +use crate::base_llm::anthropic_messages::transformation::BaseAnthropicMessagesConfig; pub struct MessagesRequest<'a> { pub model: &'a str, @@ -15,14 +15,14 @@ pub struct MessagesRequest<'a> { pub timeout: Option, } -pub(super) struct ProviderMessagesRequest { - pub(super) provider: String, - pub(super) model: String, - pub(super) config: &'static dyn BaseAnthropicMessagesConfig, - pub(super) url: String, - pub(super) body: Value, - pub(super) upstream_headers: Vec<(String, String)>, - pub(super) timeout: Option, +pub struct ProviderMessagesRequest { + pub provider: String, + pub model: String, + pub config: &'static dyn BaseAnthropicMessagesConfig, + pub url: String, + pub body: Value, + pub upstream_headers: Vec<(String, String)>, + pub timeout: Option, } #[derive(Clone, Debug, PartialEq, Serialize, Deserialize)] diff --git a/litellm-rust/crates/providers/src/provider_resolution.rs b/litellm-rust/crates/providers/src/provider_resolution.rs new file mode 100644 index 00000000000..d1ada2472e9 --- /dev/null +++ b/litellm-rust/crates/providers/src/provider_resolution.rs @@ -0,0 +1,33 @@ +#[derive(Debug, Clone, Copy, PartialEq, Eq)] +pub struct CustomLlmProvider<'a> { + pub model: &'a str, + pub custom_llm_provider: &'a str, +} + +pub fn get_custom_llm_provider<'a>( + model: &'a str, + custom_llm_provider: Option<&'a str>, +) -> Option> { + if let Some(custom_llm_provider) = custom_llm_provider.filter(|provider| !provider.is_empty()) { + return Some(CustomLlmProvider { + model: strip_custom_llm_provider_prefix(model, custom_llm_provider), + custom_llm_provider, + }); + } + + let (custom_llm_provider, model) = model.split_once('/')?; + if custom_llm_provider.is_empty() || model.is_empty() { + return None; + } + Some(CustomLlmProvider { + model, + custom_llm_provider, + }) +} + +fn strip_custom_llm_provider_prefix<'a>(model: &'a str, custom_llm_provider: &str) -> &'a str { + model + .strip_prefix(custom_llm_provider) + .and_then(|model| model.strip_prefix('/')) + .unwrap_or(model) +} diff --git a/litellm/litellm_core_utils/litellm_logging.py b/litellm/litellm_core_utils/litellm_logging.py index 40621a2f68d..bbae3021677 100644 --- a/litellm/litellm_core_utils/litellm_logging.py +++ b/litellm/litellm_core_utils/litellm_logging.py @@ -90,6 +90,7 @@ from litellm.litellm_core_utils.logging_utils import ( truncate_base64_in_messages_async, ) from litellm.litellm_core_utils.model_param_helper import ModelParamHelper +from litellm.litellm_core_utils.ptu_pricing import is_spilled_over_ptu_request from litellm.litellm_core_utils.redact_messages import ( redact_message_input_output_from_custom_logger, redact_message_input_output_from_logging, @@ -1746,8 +1747,14 @@ class Logging(LiteLLMLoggingBaseClass): if transformed_result is not None: result = transformed_result + result_hidden_params: Final = getattr(result, "_hidden_params", None) or MappingProxyType({}) + result_additional_headers: Final = ( + result_hidden_params.get("additional_headers") + if isinstance(result_hidden_params, dict) + else getattr(result_hidden_params, "additional_headers", None) + ) if isinstance(result, (BaseModel, HttpxBinaryResponseContent)) and hasattr(result, "_hidden_params"): - hidden_params: Final = getattr(result, "_hidden_params", {}) + hidden_params: Final = result_hidden_params if ( "response_cost" in hidden_params and hidden_params["response_cost"] is not None ): # use cost if already calculated @@ -1762,8 +1769,17 @@ class Logging(LiteLLMLoggingBaseClass): router_model_id = self.get_router_model_id() ## RESPONSE COST ## - custom_pricing: Final = use_custom_pricing_for_model( - litellm_params=(self.litellm_params if hasattr(self, "litellm_params") else None) + spilled_over: Final = is_spilled_over_ptu_request( + model_info=_deployment_model_info(self.litellm_params if hasattr(self, "litellm_params") else None), + response_headers=self.model_call_details.get("response_headers"), + additional_headers=result_additional_headers, + ) + custom_pricing: Final = ( + False + if spilled_over + else use_custom_pricing_for_model( + litellm_params=(self.litellm_params if hasattr(self, "litellm_params") else None) + ) ) prompt = self._prompt_for_cost_calculation() @@ -5257,6 +5273,18 @@ def _get_custom_logger_settings_from_proxy_server(callback_name: str) -> dict: return {} +def _deployment_model_info(litellm_params: dict | None) -> Mapping[str, object]: + """The router-stamped deployment model_info from whichever metadata field carries it.""" + if litellm_params is None: + return MappingProxyType({}) + for metadata_key in ("metadata", "litellm_metadata"): + if not isinstance(metadata := litellm_params.get(metadata_key), Mapping): + continue + if model_info := metadata.get("model_info"): + return model_info + return MappingProxyType({}) + + def use_custom_pricing_for_model(litellm_params: dict | None) -> bool: """ Check if the model uses custom pricing diff --git a/litellm/litellm_core_utils/ptu_pricing.py b/litellm/litellm_core_utils/ptu_pricing.py index f545ba4aa3b..80f7a822b96 100644 --- a/litellm/litellm_core_utils/ptu_pricing.py +++ b/litellm/litellm_core_utils/ptu_pricing.py @@ -14,9 +14,11 @@ from typing import Final from litellm.secret_managers.main import get_secret_bool from litellm.types.router import ModelInfo -from litellm.types.utils import CustomPricingLiteLLMParams, MirroredPricingParams +from litellm.types.utils import AzureSpillover, CustomPricingLiteLLMParams, MirroredPricingParams PTU_COST_ATTRIBUTION_ENV_VAR: Final = "LITELLM_ENABLE_PTU_COST_ATTRIBUTION" +AZURE_SPILLOVER_HEADER: Final = "x-ms-is-spilled-over" +AZURE_SPILLOVER_FROM_HEADER: Final = "x-ms-spillover-from-deployment" def is_ptu_cost_attribution_enabled() -> bool: @@ -235,3 +237,33 @@ def zeroed_ptu_pricing( ), } ) + + +def is_spilled_over_ptu_request( + model_info: Mapping[str, object], + response_headers: Mapping[str, object] | None, + additional_headers: Mapping[str, object] | None, +) -> bool: + """Whether Azure served this request from pay-as-you-go capacity, so the zeroed PTU rates must not apply.""" + if ptu_terms(model_info) is None: + return False + if not is_ptu_cost_attribution_enabled(): + return False + return azure_spillover(response_headers, additional_headers) is not None + + +def azure_spillover( + response_headers: Mapping[str, object] | None, + additional_headers: Mapping[str, object] | None, +) -> AzureSpillover | None: + """The spillover Azure reports in the response headers, else None.""" + for headers, prefix in ( + (response_headers, ""), + (additional_headers, "llm_provider-"), + ): + if headers is None or str(headers.get(f"{prefix}{AZURE_SPILLOVER_HEADER}")).lower() != "true": + continue + return AzureSpillover( + from_deployment=str(v) if (v := headers.get(f"{prefix}{AZURE_SPILLOVER_FROM_HEADER}")) is not None else None + ) + return None diff --git a/litellm/llms/azure/azure.py b/litellm/llms/azure/azure.py index 587165e6991..3cb17259b93 100644 --- a/litellm/llms/azure/azure.py +++ b/litellm/llms/azure/azure.py @@ -561,6 +561,7 @@ class AzureChatCompletion(BaseAzureLLM, BaseLLM): headers, response = self.make_sync_azure_openai_chat_completion_request( azure_client=azure_client, data=data, timeout=timeout ) + logging_obj.model_call_details["response_headers"] = headers streamwrapper: Final = CustomStreamWrapper( completion_stream=response, model=model, diff --git a/litellm/llms/bedrock/common_utils.py b/litellm/llms/bedrock/common_utils.py index cb2c70e74c8..20c8258e440 100644 --- a/litellm/llms/bedrock/common_utils.py +++ b/litellm/llms/bedrock/common_utils.py @@ -34,6 +34,7 @@ if TYPE_CHECKING: _ERROR_REQUEST_URL: Final = "https://docs.litellm.ai/docs" +_OPENAI_FAMILY_MODEL_RE: Final = re.compile(r"(^|[./])openai\.") def error_response_text(response: httpx.Response) -> str: @@ -878,9 +879,10 @@ def bedrock_model_accepts_cache_points(model: str | None) -> bool: """ Whether Converse ``cachePoint`` blocks may be sent to this model. - Bedrock rejects requests carrying cachePoint blocks for models without prompt - caching support ("You invoked an unsupported model or your request did not allow - prompt caching"), so a model whose cost-map entry does not declare + OpenAI-family models only support implicit caching and never accept explicit + ``cachePoint`` blocks. Bedrock rejects requests carrying cachePoint blocks for + models without prompt caching support ("You invoked an unsupported model or your + request did not allow prompt caching"), so a model whose cost-map entry does not declare ``supports_prompt_caching`` must not receive them. A model absent from the map (an application inference profile ARN, a model newer than the map) keeps emitting so existing caching setups never silently degrade. ``litellm.utils.supports_prompt_caching`` @@ -888,6 +890,8 @@ def bedrock_model_accepts_cache_points(model: str | None) -> bool: """ if model is None: return True + if _OPENAI_FAMILY_MODEL_RE.search(model): + return False entries: Final = tuple( entry for candidate in (model, get_bedrock_base_model(model)) diff --git a/litellm/model_prices_and_context_window_backup.json b/litellm/model_prices_and_context_window_backup.json index 96cdbcffd90..602c3ec7037 100644 --- a/litellm/model_prices_and_context_window_backup.json +++ b/litellm/model_prices_and_context_window_backup.json @@ -23788,7 +23788,7 @@ "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": false + "supports_vision": true }, "fireworks_ai/accounts/fireworks/models/mixtral-8x22b-instruct-hf": { "input_cost_per_token": 1.2e-06, @@ -24114,7 +24114,7 @@ "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": false + "supports_vision": true }, "fireworks_ai/qwen3p7-plus": { "cache_read_input_token_cost": 8e-08, @@ -69184,6 +69184,27 @@ "supports_reasoning": true, "supports_vision": true }, + "typesafe/jev-1.13.0": { + "input_cost_per_token": 4.2e-08, + "litellm_provider": "typesafe", + "mode": "evaluation", + "output_cost_per_token": 0.0, + "source": "https://docs.typesafe.ai/models" + }, + "typesafe/jev-latest": { + "input_cost_per_token": 4.2e-08, + "litellm_provider": "typesafe", + "mode": "evaluation", + "output_cost_per_token": 0.0, + "source": "https://docs.typesafe.ai/models" + }, + "typesafe/jev-preview": { + "input_cost_per_token": 4.2e-08, + "litellm_provider": "typesafe", + "mode": "evaluation", + "output_cost_per_token": 0.0, + "source": "https://docs.typesafe.ai/models" + }, "wandb/zai-org/GLM-5.3-Flash": { "cache_read_input_token_cost": 5e-08, "input_cost_per_token": 1.5e-07, diff --git a/litellm/proxy/_lazy_features.py b/litellm/proxy/_lazy_features.py index faf95397fa5..2a28ea3763f 100644 --- a/litellm/proxy/_lazy_features.py +++ b/litellm/proxy/_lazy_features.py @@ -208,6 +208,7 @@ LAZY_FEATURES: Final[tuple[LazyFeature, ...]] = ( "/nvidia_nim/", "/openai/", "/openai_passthrough/", + "/typesafe/", "/vertex-ai/", "/vertex_ai/", "/vllm/", diff --git a/litellm/proxy/_lazy_openapi_snapshot.json b/litellm/proxy/_lazy_openapi_snapshot.json index 74f38b3ca6d..8a880b5832f 100644 --- a/litellm/proxy/_lazy_openapi_snapshot.json +++ b/litellm/proxy/_lazy_openapi_snapshot.json @@ -20373,6 +20373,96 @@ ] } }, + "/typesafe/{endpoint}": { + "get": { + "description": "[Docs](https://docs.litellm.ai/docs/pass_through/typesafe)", + "operationId": "typesafe_proxy_route_typesafe__endpoint__get", + "parameters": [ + { + "in": "path", + "name": "endpoint", + "required": true, + "schema": { + "title": "Endpoint", + "type": "string" + } + } + ], + "responses": { + "200": { + "content": { + "application/json": { + "schema": {} + } + }, + "description": "Successful Response" + }, + "422": { + "content": { + "application/json": { + "schema": { + "$ref": "#/components/schemas/HTTPValidationError" + } + } + }, + "description": "Validation Error" + } + }, + "security": [ + { + "APIKeyHeader": [] + } + ], + "summary": "Typesafe Proxy Route", + "tags": [ + "llm_passthrough" + ] + }, + "post": { + "description": "[Docs](https://docs.litellm.ai/docs/pass_through/typesafe)", + "operationId": "typesafe_proxy_route_typesafe__endpoint__post", + "parameters": [ + { + "in": "path", + "name": "endpoint", + "required": true, + "schema": { + "title": "Endpoint", + "type": "string" + } + } + ], + "responses": { + "200": { + "content": { + "application/json": { + "schema": {} + } + }, + "description": "Successful Response" + }, + "422": { + "content": { + "application/json": { + "schema": { + "$ref": "#/components/schemas/HTTPValidationError" + } + } + }, + "description": "Validation Error" + } + }, + "security": [ + { + "APIKeyHeader": [] + } + ], + "summary": "Typesafe Proxy Route", + "tags": [ + "llm_passthrough" + ] + } + }, "/vertex_ai/discovery/{endpoint}": { "delete": { "description": "Call any vertex discovery endpoint using the proxy.\n\nJust use `{PROXY_BASE_URL}/vertex_ai/discovery/{endpoint:path}`\n\nTarget url: `https://discoveryengine.googleapis.com`", @@ -38680,8 +38770,7 @@ "required": false, "schema": { "default": 10, - "maximum": 100, - "minimum": 1, + "minimum": 0, "title": "Count", "type": "integer" } @@ -39385,8 +39474,7 @@ "required": false, "schema": { "default": 10, - "maximum": 100, - "minimum": 1, + "minimum": 0, "title": "Count", "type": "integer" } diff --git a/litellm/proxy/_types.py b/litellm/proxy/_types.py index 680c63393e8..b0d31df92ce 100644 --- a/litellm/proxy/_types.py +++ b/litellm/proxy/_types.py @@ -51,6 +51,7 @@ from litellm.types.router import RouterErrors, UpdateRouterConfig from litellm.types.router_weights import validate_router_settings_dict from litellm.types.secret_managers.main import KeyManagementSystem from litellm.types.utils import ( + AzureSpillover, CallTypes, CostBreakdown, EmbeddingResponse, @@ -483,6 +484,7 @@ class LiteLLMRoutes(enum.Enum): "/eu.assemblyai", "/vllm", "/mistral", + "/typesafe", "/milvus", "/gigachat", "/watsonx", @@ -850,6 +852,7 @@ class LiteLLMRoutes(enum.Enum): "/team/member_add", "/team/member_delete", "/management/v1/teams/{team_id}/members/bulk_delete", + "/management/v1/teams/{team_id}/members/bulk_update", "/team/member_update", "/team/{team_id}/member/{user_id}/reset_spend", "/team/permissions_list", @@ -3895,6 +3898,7 @@ class SpendLogsMetadata(TypedDict): autorouter_savings: ReadOnly[float | None] # stamped by the logging payload; None = not auto-routed litellm_gateway_injected_cache: ReadOnly[str | None] router_metadata: ReadOnly[SpendLogsRouterMetadata | None] # None = deployment not flagged internal_router_model + azure_spillover: ReadOnly[AzureSpillover | None] # None = Azure did not report spillover class SpendLogsPayload(TypedDict): diff --git a/litellm/proxy/auth/litellm_license.py b/litellm/proxy/auth/litellm_license.py index 6a1090a0d3a..64608567f92 100644 --- a/litellm/proxy/auth/litellm_license.py +++ b/litellm/proxy/auth/litellm_license.py @@ -17,6 +17,7 @@ if TYPE_CHECKING: AUTO_ROUTER_LICENSE_FEATURE: Final = "auto_router" +LICENSE_ALL_FEATURES: Final = "*" AUTO_ROUTER_LICENSE_REMEDY: Final = "A LiteLLM license with the 'auto_router' feature lifts the limit." @@ -153,17 +154,21 @@ class LicenseCheck: return False return team_count > _max_teams_in_license + def grants_feature(self, feature: str) -> bool: + if self.airgapped_license_data is None: + return False + allowed_features: Final = self.airgapped_license_data.get("allowed_features") + granted: Final = allowed_features if isinstance(allowed_features, list) else (allowed_features,) + return feature in granted or LICENSE_ALL_FEATURES in granted + def auto_router_capability_limit(self) -> int | None: """ How many auto-routers may claim each gated classifier or customization capability: - unlimited (None) only when the signed license lists the auto_router - feature, otherwise one per capability. A license verified through the API carries no - feature list, so it does not lift the limit either. + unlimited (None) only when the signed license lists the auto_router feature or the + "*" wildcard that grants every feature, otherwise one per capability. A license verified + through the API carries no feature list, so it does not lift the limit either. """ - if self.airgapped_license_data is None: - return 1 - allowed_features: Final = self.airgapped_license_data.get("allowed_features") - if isinstance(allowed_features, list) and AUTO_ROUTER_LICENSE_FEATURE in allowed_features: + if self.grants_feature(AUTO_ROUTER_LICENSE_FEATURE): return None return 1 diff --git a/litellm/proxy/auth/route_checks.py b/litellm/proxy/auth/route_checks.py index 166a0500cee..0a6b618805d 100644 --- a/litellm/proxy/auth/route_checks.py +++ b/litellm/proxy/auth/route_checks.py @@ -31,6 +31,7 @@ _PROXY_ADMIN_VIEW_ONLY_BLOCKED_ROUTES: Final = frozenset( # team "/team/new", "/management/v1/teams/{team_id}/members/bulk_delete", + "/management/v1/teams/{team_id}/members/bulk_update", "/team/update", "/team/delete", "/team/block", @@ -767,6 +768,7 @@ class RouteChecks: "/user/bulk_update", "/team/new", "/management/v1/teams/{team_id}/members/bulk_delete", + "/management/v1/teams/{team_id}/members/bulk_update", "/team/update", "/team/delete", "/model/new", diff --git a/litellm/proxy/client/cli/README.md b/litellm/proxy/client/cli/README.md index c8422e270de..a02d7cce0d8 100644 --- a/litellm/proxy/client/cli/README.md +++ b/litellm/proxy/client/cli/README.md @@ -569,15 +569,15 @@ lite --base-url https://your-proxy.example.com configure claude --api-key sk-... claude ``` -The key comes from `--api-key` (or `lite --api-key` / `LITELLM_PROXY_API_KEY`) and is written into `env.ANTHROPIC_AUTH_TOKEN`; without one the command refuses, since a `lite login` credential expires within a day and keeping it fresh would mean Claude Code running `lite` through `apiKeyHelper` on every credential refresh. The command checks the key against `GET /v1/models`, then patches `~/.claude/settings.json`: `env.ANTHROPIC_BASE_URL`, the credential, and `env.ENABLE_TOOL_SEARCH` and `env.CLAUDE_CODE_ENABLE_GATEWAY_MODEL_DISCOVERY` when those are missing, so Claude Code's `/model` picker lists the proxy's models (under `claude-router-` for a group whose id contains neither `claude` nor `anthropic`, since Claude Code lists only those) and you pick between them as usual. Claude Code keeps its own default model until you switch, so that id has to exist on the proxy for the first message to go through; `--model` (or the interactive prompt below) sets the model Claude Code starts on instead, as the top-level `model` key and as `env.ANTHROPIC_MODEL`, both of which have to be on `/v1/models` for the key. The second one matters for `claude -c` and `claude --resume`: a resumed session otherwise re-sends the model its transcript recorded, which behind an auto-router with `return_raw_model_name: true` is the tier model that answered, and a key scoped to the router alias gets a 403 for it; `ANTHROPIC_MODEL` outranks the transcript on resume. Nothing forces Claude Code's sub-agent or background tiers onto a proxy model, so those built-in ids need to exist on the proxy too; `lite autoroute up` is the mode that pins every tier to one group. Claude Code treats a name it does not know as an unknown model: it prints a one-line `unrecognized_model` note, assumes a 200k context window (the proxy appends `[1m]` for a group whose configured or known input window reaches 1M) and sends no thinking parameters for it, so name the group like a Claude model id to change that. The other credential slots (`env.ANTHROPIC_API_KEY`, a stale `env.ANTHROPIC_AUTH_TOKEN` or `apiKeyHelper`) are removed so they cannot fight the one written. Every other setting is preserved and the file is written atomically with owner-only permissions; if `settings.json` is a symlink into a dotfiles repository, the key is written through to that target and the command says so, so keep it out of version control +The key comes from `--api-key` (or `lite --api-key` / `LITELLM_PROXY_API_KEY`) and is written into `env.ANTHROPIC_AUTH_TOKEN`; without one the command refuses, since a `lite login` credential expires within a day and keeping it fresh would mean Claude Code running `lite` through `apiKeyHelper` on every credential refresh. The command checks the key against `GET /v1/models`, then patches `~/.claude/settings.json`: `env.ANTHROPIC_BASE_URL`, the credential, and `env.ENABLE_TOOL_SEARCH` and `env.CLAUDE_CODE_ENABLE_GATEWAY_MODEL_DISCOVERY` when those are missing, so Claude Code's `/model` picker lists the proxy's models (under `claude-router-` for a group whose id contains neither `claude` nor `anthropic`, since Claude Code lists only those) and you pick between them as usual. Claude Code keeps its own default model until you switch, so that id has to exist on the proxy for the first message to go through; `--model` (or the interactive prompt below) sets the model Claude Code starts on instead, as the top-level `model` key and as `env.ANTHROPIC_MODEL`, both of which have to be on `/v1/models` for the key. The second one matters for `claude -c` and `claude --resume`: a resumed session otherwise re-sends the model its transcript recorded, which behind an auto-router with `return_raw_model_name: true` is the tier model that answered, and a key scoped to the router alias gets a 403 for it; `ANTHROPIC_MODEL` outranks the transcript on resume. Nothing forces Claude Code's sub-agent or background tiers onto a proxy model, so those built-in ids need to exist on the proxy too; `lite autoroute start` is the mode that pins every tier to one group. Claude Code treats a name it does not know as an unknown model: it prints a one-line `unrecognized_model` note, assumes a 200k context window (the proxy appends `[1m]` for a group whose configured or known input window reaches 1M) and sends no thinking parameters for it, so name the group like a Claude model id to change that. The other credential slots (`env.ANTHROPIC_API_KEY`, a stale `env.ANTHROPIC_AUTH_TOKEN` or `apiKeyHelper`) are removed so they cannot fight the one written. Every other setting is preserved and the file is written atomically with owner-only permissions; if `settings.json` is a symlink into a dotfiles repository, the key is written through to that target and the command says so, so keep it out of version control Plain `lite configure`, with no agent named, asks which agents to wire and which gateway model each starts on, picked from `/v1/models` with a type-to-filter prompt. All choices and selected config files are checked before the first settings write. If a later filesystem write fails, the output identifies each agent already configured and its undo command -What the command changed is recorded in `~/.litellm/claude_configure_state.json` (previous values plus fingerprints of what was written, never a second copy of the key). `lite unconfigure claude` restores each of those keys only if it still holds what `configure` wrote, so anything you changed since is left alone and named in the output; a `settings.json` or `env` object that only existed because of `configure` is removed again. Ownership moves only by a write: running `configure` again (a re-login is one) refreshes the record only for the keys its merge changed, keeps the original snapshot of a key that still holds what it wrote, and snapshots afresh a key you changed in between, so `unconfigure` brings back whatever the repeat displaced and never adopts your edit as its own. A credential (`env.ANTHROPIC_API_KEY`, `env.ANTHROPIC_AUTH_TOKEN`, `apiKeyHelper`) is put back only when the restored file points at the `ANTHROPIC_BASE_URL` it was captured next to; otherwise it stays removed, the output says which server it belonged to, and the receipt is kept so pointing the URL back and running `unconfigure` again finishes the job. It also undoes `lite login --config-claude`, which writes through the same path. Both refuse to run while a `lite up` or `lite autoroute up` session holds a backup, and that check comes before any request +What the command changed is recorded in `~/.litellm/claude_configure_state.json` (previous values plus fingerprints of what was written, never a second copy of the key). `lite unconfigure claude` restores each of those keys only if it still holds what `configure` wrote, so anything you changed since is left alone and named in the output; a `settings.json` or `env` object that only existed because of `configure` is removed again. Ownership moves only by a write: running `configure` again (a re-login is one) refreshes the record only for the keys its merge changed, keeps the original snapshot of a key that still holds what it wrote, and snapshots afresh a key you changed in between, so `unconfigure` brings back whatever the repeat displaced and never adopts your edit as its own. A credential (`env.ANTHROPIC_API_KEY`, `env.ANTHROPIC_AUTH_TOKEN`, `apiKeyHelper`) is put back only when the restored file points at the `ANTHROPIC_BASE_URL` it was captured next to; otherwise it stays removed, the output says which server it belonged to, and the receipt is kept so pointing the URL back and running `unconfigure` again finishes the job. It also undoes `lite login --config-claude`, which writes through the same path. Both refuse to run while a `lite up` or `lite autoroute start` session holds a backup, and that check comes before any request #### Routed model and savings in the status line -`lite configure claude`, `lite login --config-claude`, `lite up` and `lite autoroute up` also install a status line (`~/.litellm/statusline.py`, registered as `statusLine` in `~/.claude/settings.json` unless you already run one) that shows which model the auto-router actually served the last turn and, once the proxy has recorded the session, what the session cost against the router's savings baseline: +`lite configure claude`, `lite login --config-claude`, `lite up` and `lite autoroute start` also install a status line (`~/.litellm/statusline.py`, registered as `statusLine` in `~/.claude/settings.json` unless you already run one) that shows which model the auto-router actually served the last turn and, once the proxy has recorded the session, what the session cost against the router's savings baseline: ``` Routed to: claude-haiku-4-5 -63% vs Claude Opus 5 @@ -597,7 +597,7 @@ After upgrading the CLI, rerun your original `lite configure claude` command wit #### Install the CLI -`lite autoroute up` builds and runs a throwaway litellm proxy locally, so unlike the rest of this CLI it needs the proxy server runtime, not just the thin `litellm[cli]` client. Install `litellm[proxy]` (which ships the `lite` command too) with a single curl command -- no existing Python tooling required, `uv` is bootstrapped automatically if missing: +`lite autoroute start` builds and runs a throwaway litellm proxy locally, so unlike the rest of this CLI it needs the proxy server runtime, not just the thin `litellm[cli]` client. Install `litellm[proxy]` (which ships the `lite` command too) with a single curl command -- no existing Python tooling required, `uv` is bootstrapped automatically if missing: ```bash curl -fsSL https://raw.githubusercontent.com/BerriAI/litellm/main/scripts/install.sh | sh @@ -610,7 +610,7 @@ curl -fsSL https://raw.githubusercontent.com/BerriAI/litellm// LITELLM_CLI_REF= sh ``` -The thin `scripts/install-cli.sh` installs only `litellm[cli]`, which is enough for `lite login`, `lite claude`, and `lite up`, but not for `lite autoroute up`; running it against a `litellm[cli]` install fails fast with a message telling you to install the proxy runtime. +The thin `scripts/install-cli.sh` installs only `litellm[cli]`, which is enough for `lite login`, `lite claude`, and `lite up`, but not for `lite autoroute start`; running it against a `litellm[cli]` install fails fast with a message telling you to install the proxy runtime. Point the CLI at your real proxy and key before running any `lite model-groups` or `lite autoroute` command -- like every other command in this CLI, they read `LITELLM_PROXY_URL`/`LITELLM_PROXY_API_KEY` (or `--base-url`/`--api-key`), no `lite login` required: @@ -637,44 +637,46 @@ An interactive wizard. It runs the same model-group discovery as above, splits t The wizard writes the result to `~/.litellm/autorouter/config.yaml` with `0600` permissions, since the file embeds your real proxy API key. Every model referenced anywhere in that config -- tier targets, the classifier model, the embedding model -- becomes its own `litellm_proxy/` deployment whose `api_base` and `api_key` point back at your real proxy. That is the trick that keeps your real proxy's config untouched: every actual network call this generates, whether it is the routed completion, an LLM-classifier call, or an embedding call, forwards transparently through your real, already-running proxy with your real key. -You do not need to tell Claude Code to request `autorouter` by name yourself: `lite autoroute up` also sets the top-level `model` and `ANTHROPIC_DEFAULT_SONNET_MODEL`, `ANTHROPIC_DEFAULT_HAIKU_MODEL`, `ANTHROPIC_DEFAULT_OPUS_MODEL` and `ANTHROPIC_DEFAULT_FABLE_MODEL` to `autorouter` in `~/.claude/settings.json` (and `CLAUDE_CODE_ENABLE_GATEWAY_MODEL_DISCOVERY` to `1` when missing, like every other wiring), so every one of Claude Code's own model tiers requests it directly regardless of `/model` or whatever it defaults to otherwise. (A bare `model_name: "*"` deployment looks like the obvious way to catch any request instead, but litellm's Router looks up auto-router deployments by the literal requested model string with no wildcard resolution, so a `"*"` entry would never actually match real traffic -- these env var overrides are what makes it work.) +You do not need to tell Claude Code to request `autorouter` by name yourself: `lite autoroute start` also sets the top-level `model` and `ANTHROPIC_DEFAULT_SONNET_MODEL`, `ANTHROPIC_DEFAULT_HAIKU_MODEL`, `ANTHROPIC_DEFAULT_OPUS_MODEL` and `ANTHROPIC_DEFAULT_FABLE_MODEL` to `autorouter` in `~/.claude/settings.json` (and `CLAUDE_CODE_ENABLE_GATEWAY_MODEL_DISCOVERY` to `1` when missing, like every other wiring), so every one of Claude Code's own model tiers requests it directly regardless of `/model` or whatever it defaults to otherwise. (A bare `model_name: "*"` deployment looks like the obvious way to catch any request instead, but litellm's Router looks up auto-router deployments by the literal requested model string with no wildcard resolution, so a `"*"` entry would never actually match real traffic -- these env var overrides are what makes it work.) -You must run `configure` at least once before `up`; running `up` first fails with a clear error telling you to configure first. +You must run `configure` at least once before `start`; running `start` first fails with a clear error telling you to configure first. #### Launch the Ephemeral Auto-Router Proxy ```bash -lite autoroute up +lite autoroute start ``` -Starts a local, throwaway litellm proxy on `127.0.0.1:5483` (override with `--port`), running the config `configure` generated, with a self-issued API key baked in (your real proxy key never leaves the generated config -- it only appears there, forwarding to your real proxy). Both the port and the key are stable across runs: the key is minted once, persisted inside the generated config, and reused by every later `up` (and carried forward when you re-run `configure`), so anything you configured against one session keeps working in the next. If the port is already taken, `up` refuses with a clear error instead of silently moving to another one. It waits for the ephemeral proxy to report healthy, then patches `~/.claude/settings.json` the same way `lite up` does, except with a static `ANTHROPIC_AUTH_TOKEN` env var instead of an `apiKeyHelper`, since this key is self-issued rather than something needing SSO refresh. Any `claude` session started afterward, from any terminal, routes through the ephemeral proxy. +Starts a local, throwaway litellm proxy on `127.0.0.1:5483` (override with `--port`), running the config `configure` generated, with a self-issued API key baked in (your real proxy key never leaves the generated config -- it only appears there, forwarding to your real proxy). Both the port and the key are stable across runs: the key is minted once, persisted inside the generated config, and reused by every later `start` (and carried forward when you re-run `configure`), so anything you configured against one session keeps working in the next. If the port is already taken, `start` refuses with a clear error instead of silently moving to another one. It waits for the ephemeral proxy to report healthy, then patches `~/.claude/settings.json` the same way `lite up` does, except with a static `ANTHROPIC_AUTH_TOKEN` env var instead of an `apiKeyHelper`, since this key is self-issued rather than something needing SSO refresh. Any `claude` session started afterward, from any terminal, routes through the ephemeral proxy. -`lite autoroute up` runs in the foreground and streams the ephemeral proxy's own log file into your terminal, so you can watch its routing decisions -- which tier and model got picked for each request -- as you use Claude Code normally. Press Ctrl-C (or send SIGTERM) to stop it; this kills the child proxy process and restores your original Claude Code settings, in that order. +`lite autoroute start` runs in the foreground and streams the ephemeral proxy's own log file into your terminal, so you can watch its routing decisions -- which tier and model got picked for each request -- as you use Claude Code normally. Press Ctrl-C (or send SIGTERM) to stop it; this kills the child proxy process and restores your original Claude Code settings, in that order. #### Recover From an Unclean Shutdown ```bash -lite autoroute down +lite autoroute stop ``` -If the `lite autoroute up` process dies uncleanly -- `kill -9`, a crash -- rather than being stopped with Ctrl-C, `down` is the manual recovery path: it kills any leftover ephemeral proxy process found via a recorded pid file and restores Claude Code's settings from whatever backup is on disk. +If the `lite autoroute start` process dies uncleanly -- `kill -9`, a crash -- rather than being stopped with Ctrl-C, `stop` is the manual recovery path: it kills any leftover ephemeral proxy process found via a recorded pid file and restores Claude Code's settings from whatever backup is on disk. #### Example ```bash lite autoroute configure -lite autoroute up +lite autoroute start # use Claude Code as normal in another terminal; routing decisions stream live -lite autoroute down # only needed if `up` was killed uncleanly instead of Ctrl-C'd +lite autoroute stop # only needed if `start` was killed uncleanly instead of Ctrl-C'd ``` +The previous names, `lite autoroute up` and `lite autoroute down`, still work as hidden aliases of `start` and `stop`: each prints a deprecation notice on stderr and will be removed in a future release + #### Caveats -Adaptive mode's learned state does not persist across `lite autoroute up` sessions -- there is no local database, so every session starts adaptive selection cold. A Claude Code session already running before `up` started, or still running when it stops, keeps whatever settings it loaded at its own startup; like `lite up`, this is a one-time file patch and restore, not a live traffic interceptor. Only Claude Code is supported, for the same reason as `lite up`: no other supported agent (for example Cursor) has an equivalent hot-patchable config file. +Adaptive mode's learned state does not persist across `lite autoroute start` sessions -- there is no local database, so every session starts adaptive selection cold. A Claude Code session already running before `start` ran, or still running when it stops, keeps whatever settings it loaded at its own startup; like `lite up`, this is a one-time file patch and restore, not a live traffic interceptor. Only Claude Code is supported, for the same reason as `lite up`: no other supported agent (for example Cursor) has an equivalent hot-patchable config file. -A session that outlives `up` (or is still running the moment you stop it) keeps sending requests, master key included, to that now-freed loopback port until you restart it. Once the ephemeral proxy process exits, nothing stops another local account on the same machine from binding that same port and receiving those requests instead -- and since the port is a fixed, predictable default and the master key is a static value that persists across sessions (unlike `lite up`'s `apiKeyHelper`, which is re-resolved per request), whoever receives them gets a live-looking token along with the prompt content. Restart any Claude Code session before you consider the machine clean, run `lite autoroute down` promptly rather than leaving a stopped session's settings patched, and do not run `lite autoroute up` on a shared or multi-tenant host. To rotate the persisted key, delete the `master_key` line from `~/.litellm/autorouter/config.yaml`; the next `up` mints a fresh one (deleting the whole file works too, but then `configure` must be re-run first). +A session that outlives `start` (or is still running the moment you stop it) keeps sending requests, master key included, to that now-freed loopback port until you restart it. Once the ephemeral proxy process exits, nothing stops another local account on the same machine from binding that same port and receiving those requests instead -- and since the port is a fixed, predictable default and the master key is a static value that persists across sessions (unlike `lite up`'s `apiKeyHelper`, which is re-resolved per request), whoever receives them gets a live-looking token along with the prompt content. Restart any Claude Code session before you consider the machine clean, run `lite autoroute stop` promptly rather than leaving a stopped session's settings patched, and do not run `lite autoroute start` on a shared or multi-tenant host. To rotate the persisted key, delete the `master_key` line from `~/.litellm/autorouter/config.yaml`; the next `start` mints a fresh one (deleting the whole file works too, but then `configure` must be re-run first). -Do not run `lite up` and `lite autoroute up` at the same time. Each patches `~/.claude/settings.json` and keeps its own separate backup, with no coordination between them: whichever one you stop or crash out of last is the one whose backup gets restored, which can silently leave the *other* mode's settings (a static master key and a now-dead loopback URL, or a stale `apiKeyHelper`) active. Run `lite down` or `lite autoroute down` (whichever applies) before switching to the other mode. +Do not run `lite up` and `lite autoroute start` at the same time. Each patches `~/.claude/settings.json` and keeps its own separate backup, with no coordination between them: whichever one you stop or crash out of last is the one whose backup gets restored, which can silently leave the *other* mode's settings (a static master key and a now-dead loopback URL, or a stale `apiKeyHelper`) active. Run `lite down` or `lite autoroute stop` (whichever applies) before switching to the other mode. ## Environment Variables diff --git a/litellm/proxy/client/cli/__init__.py b/litellm/proxy/client/cli/__init__.py index 843a0095878..7634cabb3b3 100644 --- a/litellm/proxy/client/cli/__init__.py +++ b/litellm/proxy/client/cli/__init__.py @@ -1,5 +1,5 @@ """CLI package for LiteLLM Proxy Client.""" -from .main import cli +from .main import cli, litellm_proxy_cli -__all__ = ["cli"] +__all__ = ["cli", "litellm_proxy_cli"] diff --git a/litellm/proxy/client/cli/commands/autoroute/commands.py b/litellm/proxy/client/cli/commands/autoroute/commands.py index 5d91fc81350..05c21875f84 100644 --- a/litellm/proxy/client/cli/commands/autoroute/commands.py +++ b/litellm/proxy/client/cli/commands/autoroute/commands.py @@ -51,7 +51,7 @@ def _ensure_master_key() -> str: The generated config is the single home of the key: the proxy server authenticates against general_settings.master_key only (a key under litellm_settings is silently ignored, which would leave the ephemeral proxy with no real auth), and the file is written 0600 via - secure_create. Reusing that persisted value keeps the key stable across `up` runs, so a + secure_create. Reusing that persisted value keeps the key stable across `start` runs, so a client configured against one session keeps working in the next. """ with open(CONFIG_PATH, "r") as f: @@ -88,15 +88,18 @@ def configure(ctx: click.Context) -> None: run_configure_wizard(ctx) -@autoroute_group.command("up") -@click.option( +_PORT_OPTION: Final = click.option( "--port", type=click.IntRange(1, 65535), default=DEFAULT_AUTOROUTE_PORT, show_default=True, help="Loopback port for the ephemeral proxy; stable across runs so configured clients keep working.", ) -def up(port: int) -> None: + + +@autoroute_group.command("start") +@_PORT_OPTION +def start(port: int) -> None: """Launch the ephemeral auto-router proxy and route Claude Code through it""" if not CONFIG_PATH.exists(): raise click.ClickException("No config found. Run `lite autoroute configure` first.") @@ -104,7 +107,7 @@ def up(port: int) -> None: missing: Final = missing_proxy_runtime_modules() if missing: raise click.ClickException( - "lite autoroute up launches a local litellm proxy, which needs the proxy runtime that the " + "lite autoroute start launches a local litellm proxy, which needs the proxy runtime that the " f"thin `litellm[cli]` install does not include (missing: {', '.join(missing)}). Install the " "proxy runtime with `uv tool install --force 'litellm[proxy]'`, or to QA a branch, " "`curl -fsSL https://raw.githubusercontent.com/BerriAI/litellm//scripts/install.sh | " @@ -117,14 +120,14 @@ def up(port: int) -> None: raise click.ClickException(str(e)) if existing_pid is not None and is_running(existing_pid.pid): raise click.ClickException( - "An ephemeral proxy is already running (lite autoroute up looks already active). " - "Run `lite autoroute down` first." + "An ephemeral proxy is already running (lite autoroute start looks already active). " + "Run `lite autoroute stop` first." ) if AUTOROUTE_BACKUP_PATH.exists(): raise click.ClickException( - f"{AUTOROUTE_BACKUP_PATH} already exists -- `lite autoroute up` looks like it's already " - "running (or crashed without cleanup). Run `lite autoroute down` first." + f"{AUTOROUTE_BACKUP_PATH} already exists -- `lite autoroute start` looks like it's already " + "running (or crashed without cleanup). Run `lite autoroute stop` first." ) if port == 4000: @@ -135,8 +138,8 @@ def up(port: int) -> None: if not is_port_available(port): raise click.ClickException( - f"Port {port} on 127.0.0.1 is already in use. If a previous `lite autoroute up` is still " - "running or crashed, run `lite autoroute down`; otherwise pick a different port with --port." + f"Port {port} on 127.0.0.1 is already in use. If a previous `lite autoroute start` is still " + "running or crashed, run `lite autoroute stop`; otherwise pick a different port with --port." ) master_key: Final = _ensure_master_key() @@ -196,7 +199,7 @@ def up(port: int) -> None: click.echo("\nStopped ephemeral proxy and restored Claude Code settings.") click.echo( f"Restart any Claude Code session still open from this session, or another local account could " - f"bind the now-free port {port} and receive its requests. Do not use `lite autoroute up` on a " + f"bind the now-free port {port} and receive its requests. Do not use `lite autoroute start` on a " f"shared or multi-tenant host." ) @@ -214,13 +217,13 @@ def up(port: int) -> None: _teardown() -@autoroute_group.command("down") -def down() -> None: +@autoroute_group.command("stop") +def stop() -> None: """Restore Claude Code settings and stop a leftover ephemeral proxy, if any""" try: record: PidRecord | None = read_pid_record() except ClaudeSettingsError as e: - # down is the crash-recovery path -- a corrupt pid record must not block it; clear the + # stop is the crash-recovery path -- a corrupt pid record must not block it; clear the # unusable record and keep going rather than leaving the user with no way to clean up. click.echo(f"{e} Clearing it and continuing cleanup.", err=True) record = None @@ -238,7 +241,34 @@ def down() -> None: elif restored.existed: click.echo(f"Restored {CLAUDE_SETTINGS_PATH} to its original contents.") else: - click.echo(f"Removed {CLAUDE_SETTINGS_PATH} (it did not exist before `lite autoroute up`).") + click.echo(f"Removed {CLAUDE_SETTINGS_PATH} (it did not exist before `lite autoroute start`).") + + +AUTOROUTE_ALIAS_DEPRECATION_NOTICE: Final = ( + "`lite autoroute {retired}` is deprecated and will be removed in a future release; " + "run `lite autoroute {current}` instead, it takes the same options." +) + + +def _warn_deprecated_alias(retired: str, current: str) -> None: + click.secho(AUTOROUTE_ALIAS_DEPRECATION_NOTICE.format(retired=retired, current=current), err=True, fg="yellow") + + +@autoroute_group.command("up", hidden=True) +@_PORT_OPTION +@click.pass_context +def up(ctx: click.Context, port: int) -> None: + """Deprecated alias of `lite autoroute start`""" + _warn_deprecated_alias("up", "start") + ctx.invoke(start, port=port) + + +@autoroute_group.command("down", hidden=True) +@click.pass_context +def down(ctx: click.Context) -> None: + """Deprecated alias of `lite autoroute stop`""" + _warn_deprecated_alias("down", "stop") + ctx.invoke(stop) __all__ = ["autoroute_group"] diff --git a/litellm/proxy/client/cli/commands/autoroute/config.py b/litellm/proxy/client/cli/commands/autoroute/config.py index 1f3ad34e3d9..1bfcdf444bf 100644 --- a/litellm/proxy/client/cli/commands/autoroute/config.py +++ b/litellm/proxy/client/cli/commands/autoroute/config.py @@ -214,7 +214,7 @@ def build_generated_proxy_config(config: AutorouteConfig, master_key: str) -> di def master_key_from_config(config: dict[str, JsonValue]) -> str | None: """The master key persisted in a generated config, or None when absent or blank. - Single definition of "this config already has a usable key", shared by `up` (reuse + Single definition of "this config already has a usable key", shared by `start` (reuse instead of minting) and the configure wizard (carry the key forward on rewrite) so the two sites can never disagree on what counts as one. Returned verbatim, never stripped: the proxy authenticates against the exact bytes under general_settings.master_key, so a diff --git a/litellm/proxy/client/cli/commands/autoroute/process.py b/litellm/proxy/client/cli/commands/autoroute/process.py index 425b8581fed..3d3793ec95b 100644 --- a/litellm/proxy/client/cli/commands/autoroute/process.py +++ b/litellm/proxy/client/cli/commands/autoroute/process.py @@ -43,12 +43,12 @@ _PROXY_RUNTIME_MODULES: tuple[str, ...] = ("fastapi", "uvicorn", "backoff", "orj def missing_proxy_runtime_modules() -> tuple[str, ...]: - """Proxy-server modules that ``lite autoroute up`` needs but the thin CLI install lacks. + """Proxy-server modules that ``lite autoroute start`` needs but the thin CLI install lacks. ``launch_proxy`` runs the full ``litellm.proxy.proxy_cli`` server, whose dependencies live in the ``proxy`` extra, not the ``cli`` extra that installs the ``lite`` command. On a thin ``litellm[cli]`` install the subprocess dies with a bare ``ModuleNotFoundError``; detecting the - gap here lets ``up`` fail with an actionable message instead. + gap here lets ``start`` fail with an actionable message instead. """ return tuple(name for name in _PROXY_RUNTIME_MODULES if importlib.util.find_spec(name) is None) diff --git a/litellm/proxy/client/cli/commands/autoroute/wizard.py b/litellm/proxy/client/cli/commands/autoroute/wizard.py index 11fbd5c1402..a7fd92b9e84 100644 --- a/litellm/proxy/client/cli/commands/autoroute/wizard.py +++ b/litellm/proxy/client/cli/commands/autoroute/wizard.py @@ -94,7 +94,7 @@ def _load_persisted_master_key(config_path: Path) -> str | None: """The master key from an existing generated config, so a rewrite carries it forward. Lenient on a missing or corrupt file: configure is the regeneration path, so it must - succeed from any prior state; a key that cannot be read is simply not carried and `up` + succeed from any prior state; a key that cannot be read is simply not carried and `start` mints a fresh one. """ if not config_path.exists(): diff --git a/litellm/proxy/client/cli/commands/claude_settings.py b/litellm/proxy/client/cli/commands/claude_settings.py index 1473e40070f..f4bebc4a4cb 100644 --- a/litellm/proxy/client/cli/commands/claude_settings.py +++ b/litellm/proxy/client/cli/commands/claude_settings.py @@ -1,6 +1,6 @@ """Shared handling of Claude Code's ~/.claude/settings.json. -`lite up` and `lite autoroute up` patch this file temporarily and restore it on +`lite up` and `lite autoroute start` patch this file temporarily and restore it on exit; `lite configure claude` patches it persistently and records how to undo it. All of them need the same merge, and `up` already imports from `auth`, so the shared parts live here rather than in any one command module. The credential is @@ -88,7 +88,7 @@ class SettingsFileOwner: SETTINGS_FILE_OWNERS: Final = ( SettingsFileOwner(BACKUP_PATH, "lite up", "lite down"), - SettingsFileOwner(AUTOROUTE_BACKUP_PATH, "lite autoroute up", "lite autoroute down"), + SettingsFileOwner(AUTOROUTE_BACKUP_PATH, "lite autoroute start", "lite autoroute stop"), ) _SETTINGS_ADAPTER: Final = TypeAdapter(dict[str, JsonValue]) @@ -111,7 +111,7 @@ def _is_default_settings_file(settings_path: Path) -> bool: def settings_file_owners(settings_path: Path) -> tuple[SettingsFileOwner, ...]: - """The commands whose backups guard settings_path: `lite up` and `lite autoroute up` only ever manage the default file.""" + """The commands whose backups guard settings_path: `lite up` and `lite autoroute start` only ever manage the default file.""" return SETTINGS_FILE_OWNERS if _is_default_settings_file(settings_path) else () @@ -240,7 +240,7 @@ def _env_object(settings: Mapping[str, JsonValue], path: Path) -> Mapping[str, J def refuse_while_owned(settings_path: Path, owners: Sequence[SettingsFileOwner]) -> None: - """Refuse while `lite up` or `lite autoroute up` holds a backup it will restore over any write; a + """Refuse while `lite up` or `lite autoroute start` holds a backup it will restore over any write; a purely local check, so commands run it before any login prompt or request.""" for owner in owners: if owner.backup_path.exists(): @@ -262,7 +262,7 @@ def _write_target(settings_path: Path) -> Path: def write_claude_settings(settings_path: Path, settings: Mapping[str, JsonValue]) -> None: """The one way a settings document lands on disk: staged owner-only beside the target and renamed into - place, through a symlink rather than over it. Every writer (`configure`, `up`, `autoroute up` and the + place, through a symlink rather than over it. Every writer (`configure`, `up`, `autoroute start` and the restores) may be carrying the credential, so none creates the file under the umask or truncates it.""" target: Final = _write_target(settings_path) try: @@ -341,7 +341,7 @@ def merge_claude_settings( an apiKeyHelper) are removed, since Claude Code given two credentials may send the wrong one. ENABLE_TOOL_SEARCH and CLAUDE_CODE_ENABLE_GATEWAY_MODEL_DISCOVERY get their defaults only when missing. `default_model` is the top-level `model` and env.ANTHROPIC_MODEL (see StartOn); - `tier_model` is `lite autoroute up`'s knob that points every ANTHROPIC_DEFAULT_*_MODEL at one + `tier_model` is `lite autoroute start`'s knob that points every ANTHROPIC_DEFAULT_*_MODEL at one group. Apart from those tier keys, exactly OWNED_PATHS are touched. """ raw_env: Final = settings.get(ENV_KEY, {}) diff --git a/litellm/proxy/client/cli/commands/configure.py b/litellm/proxy/client/cli/commands/configure.py index 7988f8aef3c..2878ae0e9f8 100644 --- a/litellm/proxy/client/cli/commands/configure.py +++ b/litellm/proxy/client/cli/commands/configure.py @@ -56,7 +56,7 @@ _CLAUDE_CODE_VIEW: Final = MappingProxyType( _MODEL_OPTION_HELP: Final = ( f"Proxy model to set as {STARTING_MODEL_ROLE}. Must be listed on /v1/models for the key; without it, " "Claude Code keeps its own default and a pin an earlier configure made is let go of. Nothing pins Claude " - "Code's sub-agent or background tiers; `lite autoroute up` is the mode that does." + "Code's sub-agent or background tiers; `lite autoroute start` is the mode that does." ) diff --git a/litellm/proxy/client/cli/commands/encryption.py b/litellm/proxy/client/cli/commands/encryption.py index 4c6ab94191e..f9a9356d0d6 100644 --- a/litellm/proxy/client/cli/commands/encryption.py +++ b/litellm/proxy/client/cli/commands/encryption.py @@ -36,8 +36,8 @@ def migrate(ctx: click.Context, check_only: bool, dry_run: bool): resumable; safe to re-run after an interruption. Examples: - litellm-proxy encryption migrate --check # attestation scan, no writes - litellm-proxy encryption migrate # perform the migration + lite encryption migrate --check # attestation scan, no writes + lite encryption migrate # perform the migration """ client: Final = HTTPClient(ctx.obj["base_url"], ctx.obj["api_key"]) diff --git a/litellm/proxy/client/cli/main.py b/litellm/proxy/client/cli/main.py index 05fb877d0f1..63e38c93221 100644 --- a/litellm/proxy/client/cli/main.py +++ b/litellm/proxy/client/cli/main.py @@ -168,5 +168,16 @@ cli.add_command(configure_group) cli.add_command(unconfigure_group) +LITELLM_PROXY_DEPRECATION_NOTICE: Final = ( + "The `litellm-proxy` command is deprecated and will be removed in a future release; " + "run `lite` instead, it takes the same commands and options." +) + + +def litellm_proxy_cli() -> None: + click.secho(LITELLM_PROXY_DEPRECATION_NOTICE, err=True, fg="yellow") + cli() + + if __name__ == "__main__": cli() diff --git a/litellm/proxy/common_utils/timezone_utils.py b/litellm/proxy/common_utils/timezone_utils.py index a50daf40144..99e89210e43 100644 --- a/litellm/proxy/common_utils/timezone_utils.py +++ b/litellm/proxy/common_utils/timezone_utils.py @@ -78,3 +78,27 @@ def get_budget_reset_time(budget_duration: str) -> datetime: `BudgetResetSettings` by injection (creation/update endpoints, startup backfill). """ return compute_budget_reset_at(budget_duration, get_budget_reset_settings()) + + +def _is_persistable_budget_duration(budget_duration: str) -> bool: + from litellm.litellm_core_utils.duration_parser import duration_in_seconds + + try: + if duration_in_seconds(budget_duration) <= 0: + return False + get_budget_reset_time(budget_duration=budget_duration) + except (ValueError, OverflowError): + return False + return True + + +def budget_duration_error(budget_duration: str | None) -> str | None: + """Why `budget_duration` cannot be persisted, or None when it is usable. + + A non-positive duration resolves to a reset time of "now", which leaves the row + permanently due: the reset job re-reads it every tick and, once enough of them + exist, they fill each batch and starve every other tenant's reset. + """ + if budget_duration is None or _is_persistable_budget_duration(budget_duration): + return None + return f"Invalid budget_duration '{budget_duration}'. Use a format like '1h', '24h', '7d', or '30d'." diff --git a/litellm/proxy/hooks/responses_id_security.py b/litellm/proxy/hooks/responses_id_security.py index 7e7f70d6f7e..d9050489095 100644 --- a/litellm/proxy/hooks/responses_id_security.py +++ b/litellm/proxy/hooks/responses_id_security.py @@ -22,7 +22,7 @@ from litellm.types.llms.openai import ( BaseLiteLLMOpenAIResponseObject, ResponsesAPIResponse, ) -from litellm.types.utils import CallTypesLiteral, LLMResponseTypes, SpecialEnums +from litellm.types.utils import ADDRESSED_RESPONSE_ID_FIELD, CallTypesLiteral, LLMResponseTypes, SpecialEnums if TYPE_CHECKING: from litellm.caching.caching import DualCache @@ -32,7 +32,6 @@ if TYPE_CHECKING: _RESPONSES_API_PROVIDER_PREFIX: Final = "/openai" _RESPONSES_API_CREATE_ROUTES: Final = frozenset({"/v1/responses", "/responses"}) -_ADDRESSED_RESPONSE_ID_KEY: Final = "_litellm_addressed_response_id" _UNMANAGED_RESPONSE_ID_DETAIL: Final = ( "Forbidden. This response id was not issued by this proxy, so the proxy cannot tell who owns it. " "To let keys address responses this proxy did not issue, set " @@ -132,7 +131,7 @@ class ResponsesIDSecurity(CustomLogger): if call_type not in responses_api_call_types: return None addressed_id_field: Final = "previous_response_id" if call_type == "aresponses" else "response_id" - retained_id: Final = data.get(_ADDRESSED_RESPONSE_ID_KEY) + retained_id: Final = data.get(ADDRESSED_RESPONSE_ID_FIELD) addressed_id: Final = ( retained_id if isinstance(retained_id, str) and retained_id else data.get(addressed_id_field) ) @@ -140,7 +139,7 @@ class ResponsesIDSecurity(CustomLogger): return data authorized_id: Final = self._authorize_response_id(addressed_id, user_api_key_dict) data[addressed_id_field] = authorized_id - data[_ADDRESSED_RESPONSE_ID_KEY] = addressed_id + data[ADDRESSED_RESPONSE_ID_FIELD] = addressed_id return data def _authorize_response_id( diff --git a/litellm/proxy/image_endpoints/endpoints.py b/litellm/proxy/image_endpoints/endpoints.py index 5b90c0ff830..b9580ba3948 100644 --- a/litellm/proxy/image_endpoints/endpoints.py +++ b/litellm/proxy/image_endpoints/endpoints.py @@ -36,6 +36,10 @@ router: Final = APIRouter() IMAGE_EDIT_NUMERIC_FORM_FIELDS: Final = numeric_form_fields(get_type_hints(ImageEditRequestParams)) +IMAGE_ARRAY_FIELD: Final = "image[]" +MASK_ARRAY_FIELD: Final = "mask[]" +BRACKETED_FILE_FIELDS: Final = frozenset({IMAGE_ARRAY_FIELD, MASK_ARRAY_FIELD}) + async def uploadfile_to_bytesio(upload: UploadFile) -> io.BytesIO: """ @@ -244,9 +248,9 @@ async def image_edit_api( fastapi_response: Response, user_api_key_dict: UserAPIKeyAuth = Depends(user_api_key_auth), image: list[UploadFile] | None = File(None), - image_array: list[UploadFile] | None = File(None, alias="image[]"), + image_array: list[UploadFile] | None = File(None, alias=IMAGE_ARRAY_FIELD), mask: list[UploadFile] | None = File(None), - mask_array: list[UploadFile] | None = File(None, alias="mask[]"), + mask_array: list[UploadFile] | None = File(None, alias=MASK_ARRAY_FIELD), model: str | None = None, ): """ @@ -294,12 +298,14 @@ async def image_edit_api( ######################################################### # Read request body and convert UploadFiles to BytesIO ######################################################### - data: Final = dict( - coerce_numeric_form_fields( + data: Final = { + key: value + for key, value in coerce_numeric_form_fields( parsed_body=await _read_request_body(request=request), numeric_fields=IMAGE_EDIT_NUMERIC_FORM_FIELDS, - ) - ) + ).items() + if key not in BRACKETED_FILE_FIELDS + } image_files: Final = await batch_to_bytesio(image) mask_files: Final = await batch_to_bytesio(mask) if image_files: diff --git a/litellm/proxy/management_endpoints/common_utils.py b/litellm/proxy/management_endpoints/common_utils.py index 973311608ed..29f24d2465f 100644 --- a/litellm/proxy/management_endpoints/common_utils.py +++ b/litellm/proxy/management_endpoints/common_utils.py @@ -1,5 +1,6 @@ import math from collections.abc import Mapping +from types import MappingProxyType from typing import TYPE_CHECKING, Any, Final, Optional, Union from fastapi import HTTPException, status @@ -33,23 +34,11 @@ def validate_budget_duration(budget_duration: str | None, status_code: int = 400 enough of them exist, they fill each batch and starve every other tenant's reset. """ - if budget_duration is None: - return + from litellm.proxy.common_utils.timezone_utils import budget_duration_error - from litellm.litellm_core_utils.duration_parser import duration_in_seconds - from litellm.proxy.common_utils.timezone_utils import get_budget_reset_time - - try: - if duration_in_seconds(budget_duration) <= 0: - raise ValueError("budget_duration must be positive") - get_budget_reset_time(budget_duration=budget_duration) - except (ValueError, OverflowError): - raise HTTPException( - status_code=status_code, - detail={ - "error": f"Invalid budget_duration '{budget_duration}'. Use a format like '1h', '24h', '7d', or '30d'." - }, - ) + error: Final = budget_duration_error(budget_duration) + if error is not None: + raise HTTPException(status_code=status_code, detail={"error": error}) from litellm._logging import verbose_proxy_logger @@ -490,6 +479,33 @@ _TEAM_MEMBER_BUDGET_LIMIT_FIELDS: Final = ( ) +MEMBER_BUDGET_PATCH_FIELDS: Final = MappingProxyType( + { + "max_budget_in_team": "max_budget", + "tpm_limit": "tpm_limit", + "rpm_limit": "rpm_limit", + "budget_duration": "budget_duration", + "allowed_models": "allowed_models", + } +) + + +def _prisma_value(value: object) -> object: + return list(value) if isinstance(value, tuple) else value + + +def member_budget_patch(source: BaseModel) -> dict[str, Any]: + """Map the per-member limit fields a request actually set to their budget-table + columns (merge-patch: a sent value updates, an explicit null clears, an absent + field is left untouched).""" + provided: Final = source.model_dump(exclude_unset=True) + return { + column: _prisma_value(provided[request_field]) + for request_field, column in MEMBER_BUDGET_PATCH_FIELDS.items() + if request_field in provided + } + + def _is_set_budget_value(value: object) -> bool: if value is None: return False @@ -513,6 +529,7 @@ async def _upsert_budget_and_membership( user_api_key_dict: UserAPIKeyAuth, budget_patch: dict[str, Any], team_default_budget_id: str | None = None, + shared_budget_ids: frozenset[str] | None = None, ): """ Apply a merge-patch of per-member budget fields to a team membership. @@ -527,6 +544,10 @@ async def _upsert_budget_and_membership( (from team metadata.team_member_budget_id). When the membership still points at it, we clone-on-write so editing one member's budget does not mutate the shared default that every other member points at. + + ``shared_budget_ids`` extends that protection to any other row more than one + membership points at, which a caller patching several members at once has + already counted; a row listed there is cloned rather than written in place. """ if not budget_patch: return @@ -538,10 +559,8 @@ async def _upsert_budget_and_membership( get_budget_reset_time(budget_duration=duration) if duration is not None else None ) - is_shared_default: Final = ( - existing_budget_id is not None - and team_default_budget_id is not None - and existing_budget_id == team_default_budget_id + is_shared_default: Final = existing_budget_id is not None and ( + existing_budget_id == team_default_budget_id or existing_budget_id in (shared_budget_ids or frozenset()) ) async def _disconnect(): @@ -563,25 +582,25 @@ async def _upsert_budget_and_membership( ) return - create_data: Final[dict[str, Any]] = { + source_row: Final = ( + await tx.litellm_budgettable.find_unique(where={"budget_id": existing_budget_id}) if is_shared_default else None + ) + source: Final[Mapping[str, Any]] = source_row.model_dump() if source_row is not None else MappingProxyType({}) + + create_data: Final[dict[str, Any]] = { # mutable-ok: Prisma create payloads are dict-shaped "created_by": user_api_key_dict.user_id or "", "updated_by": user_api_key_dict.user_id or "", + **MappingProxyType( + {f: source[f] for f in _TEAM_MEMBER_BUDGET_LIMIT_FIELDS if _is_set_budget_value(source.get(f))} + ), + **write_data, } - if is_shared_default: - default_budget_row: Final = await tx.litellm_budgettable.find_unique(where={"budget_id": existing_budget_id}) - if default_budget_row is not None: - default_budget_dict: Final = default_budget_row.model_dump() - for field in _TEAM_MEMBER_BUDGET_LIMIT_FIELDS: - value = default_budget_dict.get(field) - if _is_set_budget_value(value): - create_data[field] = value - - create_data.update(write_data) - - if create_data.get("budget_duration") is not None: - create_data["budget_reset_at"] = get_budget_reset_time(budget_duration=create_data["budget_duration"]) - else: + # Restarting an inherited window on an unrelated edit hands the member a free period. + carried: Final = source.get("budget_reset_at") if "budget_duration" not in budget_patch else None + if carried is not None: + create_data["budget_reset_at"] = carried + if create_data.get("budget_reset_at") is None: create_data.pop("budget_reset_at", None) if not _has_meaningful_budget_limit(create_data): diff --git a/litellm/proxy/management_endpoints/management_v1/teams.py b/litellm/proxy/management_endpoints/management_v1/teams.py index ba384bfb028..eab641b2a27 100644 --- a/litellm/proxy/management_endpoints/management_v1/teams.py +++ b/litellm/proxy/management_endpoints/management_v1/teams.py @@ -1,20 +1,23 @@ -"""`POST /management/v1/teams/{team_id}/members/bulk_delete`.""" +"""`POST /management/v1/teams/{team_id}/members/bulk_delete` and `.../members/bulk_update`.""" from typing import Annotated, Final -from fastapi import APIRouter, Depends +from fastapi import APIRouter, Depends, Header from litellm._logging import verbose_proxy_logger from litellm.proxy._types import CommonProxyErrors, UserAPIKeyAuth from litellm.proxy.auth.user_api_key_auth import user_api_key_auth from litellm.proxy.list_api.common import PROBLEM_TYPE_BASE, ManagementProblem, reject_unknown_query_params from litellm.proxy.management_endpoints.management_v1.common import MANAGEMENT_V1_PREFIX +from litellm.proxy.management_helpers.bulk_team_member_budgets import bulk_update_team_member_budgets from litellm.proxy.management_helpers.bulk_user_deletion import bulk_remove_team_members from litellm.proxy.management_helpers.utils import ( management_endpoint_wrapper, # pyright: ignore[reportUnknownVariableType] # legacy decorator is untyped ) from litellm.types.proxy.management_endpoints.management_v1 import ProblemDetail from litellm.types.proxy.management_endpoints.team_endpoints import ( + BulkTeamMemberBudgetUpdateRequest, + BulkTeamMemberBudgetUpdateResponse, BulkTeamMemberDeleteRequest, BulkTeamMemberDeleteResponse, ) @@ -92,3 +95,88 @@ async def bulk_delete_team_members_action( detail="Failed to remove team members.", ) ) + + +@router.post( + "/teams/{team_id}/members/bulk_update", + tags=["team management"], # mutable-ok: FastAPI types `tags` as list[str], not Sequence + dependencies=(Depends(user_api_key_auth), Depends(reject_unknown_query_params)), + response_model=BulkTeamMemberBudgetUpdateResponse, +) +@management_endpoint_wrapper +async def bulk_update_team_member_budgets_action( + team_id: str, + data: BulkTeamMemberBudgetUpdateRequest, + user_api_key_dict: Annotated[UserAPIKeyAuth, Depends(user_api_key_auth)], + litellm_changed_by: Annotated[ + str | None, + Header( + description="The litellm-changed-by header enables tracking of actions performed by authorized users on behalf of other users, providing an audit trail for accountability", + ), + ] = None, +) -> BulkTeamMemberBudgetUpdateResponse: + """ + Set per-member limits for up to 500 members of one team in one call. Same + authorization and member addressing as `/team/member_update`: proxy admins, the team's + admins, and admins of the team's organization, with each member named by exactly one of + `user_id` or `user_email`. Unknown body fields are a 422 and an unknown team is a 404. + + Each row is a merge patch of that member's limits: a field left out is untouched, a + field sent as null is cleared, and clearing the last limit drops the member back to the + team default. A budget row shared by several memberships, the team default included, is + copied for the member being patched rather than written in place, so one member's new + cap never lands on anybody else. + + `data` holds one result per requested member, in request order, carrying the limits in + force after the write. A row is `success: false` with an `error` when it names nobody on + the team or repeats an earlier row. Roles are not part of this route; `/team/member_update` + still owns them. + + Example curl: + ``` + curl --location 'http://0.0.0.0:4000/management/v1/teams/team-1/members/bulk_update' \ + --header 'Authorization: Bearer sk-1234' \ + --header 'Content-Type: application/json' \ + --data '{"members": [{"user_id": "user-1", "max_budget_in_team": 10}, {"user_email": "user-2@example.com", "max_budget_in_team": 10, "budget_duration": "30d"}]}' + ``` + """ + try: + from litellm.proxy.proxy_server import litellm_proxy_admin_name, prisma_client, user_api_key_cache + + if prisma_client is None: + raise ManagementProblem( + ProblemDetail( + type=f"{PROBLEM_TYPE_BASE}database-not-connected", + title="Database not connected", + status=503, + detail=CommonProxyErrors.db_not_connected_error.value, + ) + ) + + results: Final = await bulk_update_team_member_budgets( + team_id=team_id, + data=data, + user_api_key_dict=user_api_key_dict, + prisma_client=prisma_client, + user_api_key_cache=user_api_key_cache, + litellm_proxy_admin_name=litellm_proxy_admin_name, + litellm_changed_by=litellm_changed_by, + ) + return BulkTeamMemberBudgetUpdateResponse(data=results) + + except ManagementProblem: + raise + except Exception as e: # noqa: BLE001 # a driver error answers as a problem document, not the OpenAI error shape + verbose_proxy_logger.exception( + "litellm.proxy.management_endpoints.management_v1.teams.bulk_update_team_member_budgets_action(): " + "Exception occured - %s", + e, + ) + raise ManagementProblem( + ProblemDetail( + type=f"{PROBLEM_TYPE_BASE}internal-server-error", + title="Internal server error", + status=500, + detail="Failed to update team member budgets.", + ) + ) diff --git a/litellm/proxy/management_endpoints/scim/scim_v2.py b/litellm/proxy/management_endpoints/scim/scim_v2.py index ceb67e3eee8..34c1ad42435 100644 --- a/litellm/proxy/management_endpoints/scim/scim_v2.py +++ b/litellm/proxy/management_endpoints/scim/scim_v2.py @@ -264,6 +264,8 @@ scim_router: Final = APIRouter( dependencies=[Depends(_premium_user_check)], ) +SCIM_MAX_PAGE_SIZE: Final = 100 + # Helper functions for common operations async def _get_prisma_client_or_raise_exception(): @@ -1572,12 +1574,13 @@ def _parse_scim_eq_filter(scim_filter: str) -> tuple[str, str] | None: ) async def get_users( startIndex: int = Query(1, ge=1), - count: int = Query(10, ge=1, le=100), + count: int = Query(10, ge=0), filter: str | None = Query(None), ): """ Get a list of users according to SCIM v2 protocol """ + page_size: Final = min(count, SCIM_MAX_PAGE_SIZE) verbose_proxy_logger.debug( "SCIM GET USERS request: startIndex=%s count=%s filter=%s", startIndex, @@ -1607,7 +1610,7 @@ async def get_users( users: Final[Sequence[LiteLLM_UserTable]] = await _table(UserRepository(prisma_client)).find_many( where=where_conditions, skip=(startIndex - 1), - take=count, + take=page_size, order={"created_at": "desc"}, ) @@ -1623,7 +1626,7 @@ async def get_users( return SCIMListResponse( totalResults=total_count, startIndex=startIndex, - itemsPerPage=min(count, len(scim_users)), + itemsPerPage=len(scim_users), Resources=scim_users, ) @@ -2399,12 +2402,13 @@ class _TeamWhereConditions(TypedDict, total=False): ) async def get_groups( startIndex: int = Query(1, ge=1), - count: int = Query(10, ge=1, le=100), + count: int = Query(10, ge=0), filter: str | None = Query(None), ): """ Get a list of groups according to SCIM v2 protocol """ + page_size: Final = min(count, SCIM_MAX_PAGE_SIZE) verbose_proxy_logger.debug( "SCIM GET GROUPS request: startIndex=%s count=%s filter=%s", startIndex, @@ -2425,7 +2429,7 @@ async def get_groups( teams: Final = await _table(TeamRepository(prisma_client)).find_many( where=where_conditions, skip=(startIndex - 1), - take=count, + take=page_size, order={"created_at": "desc"}, ) @@ -2462,7 +2466,7 @@ async def get_groups( return SCIMListResponse( totalResults=total_count, startIndex=startIndex, - itemsPerPage=min(count, len(scim_groups)), + itemsPerPage=len(scim_groups), Resources=scim_groups, ) diff --git a/litellm/proxy/management_endpoints/team_endpoints.py b/litellm/proxy/management_endpoints/team_endpoints.py index d16fc0fb40c..216480e298b 100644 --- a/litellm/proxy/management_endpoints/team_endpoints.py +++ b/litellm/proxy/management_endpoints/team_endpoints.py @@ -129,6 +129,7 @@ from litellm.proxy.management_endpoints.common_utils import ( _update_metadata_fields, _upsert_budget_and_membership, _user_has_admin_view, + member_budget_patch, validate_budget_duration, validate_team_model_max_budget, ) @@ -3686,27 +3687,6 @@ async def team_member_delete( return existing_team_row -_MEMBER_BUDGET_PATCH_FIELDS: Final = { - "max_budget_in_team": "max_budget", - "tpm_limit": "tpm_limit", - "rpm_limit": "rpm_limit", - "budget_duration": "budget_duration", - "allowed_models": "allowed_models", -} - - -def _build_member_budget_patch(data: TeamMemberUpdateRequest) -> dict[str, object]: - """Map the budget fields the request actually set (merge-patch: a sent - value updates, an explicit null clears, an absent field is left untouched) - to their budget-table columns.""" - provided: Final = data.model_dump(exclude_unset=True) - return { - column: provided[request_field] - for request_field, column in _MEMBER_BUDGET_PATCH_FIELDS.items() - if request_field in provided - } - - @router.post( "/team/member_update", tags=["team management"], @@ -3812,7 +3792,7 @@ async def team_member_update( team_default_budget_id = raw_default_budget_id ### upsert new budget - budget_patch: Final = _build_member_budget_patch(data) + budget_patch: Final = member_budget_patch(data) async with prisma_client.tx() as tx: await _upsert_budget_and_membership( tx=tx, diff --git a/litellm/proxy/management_endpoints/ui_sso.py b/litellm/proxy/management_endpoints/ui_sso.py index 329443148a2..00cf357d89d 100644 --- a/litellm/proxy/management_endpoints/ui_sso.py +++ b/litellm/proxy/management_endpoints/ui_sso.py @@ -354,7 +354,7 @@ def _get_cli_sso_flow_or_raise(login_id: str | None, cache: DualCache) -> dict: status_code=400, detail=( "Your litellm CLI is out of date and uses a login flow this proxy no longer supports. " - "Upgrade it with `pip install -U 'litellm[proxy]'` and run `litellm-proxy login` again." + "Upgrade it with `pip install -U 'litellm[proxy]'` and run `lite login` again." ), ) if not _is_valid_cli_sso_login_id(login_id): @@ -375,7 +375,7 @@ def _get_cli_sso_flow_or_raise(login_id: str | None, cache: DualCache) -> dict: raise HTTPException( status_code=400, detail=( - "CLI login session not found or expired. Run `litellm-proxy login` again. " + "CLI login session not found or expired. Run `lite login` again. " "If this happens immediately after starting a login, the proxy is likely running multiple " "replicas without a shared cache; configure a Redis cache " "so every replica can see the login session." diff --git a/litellm/proxy/management_helpers/bulk_team_member_budgets.py b/litellm/proxy/management_helpers/bulk_team_member_budgets.py new file mode 100644 index 00000000000..8ca27d8d9ce --- /dev/null +++ b/litellm/proxy/management_helpers/bulk_team_member_budgets.py @@ -0,0 +1,263 @@ +"""Batched per-member limit writes behind `POST /management/v1/teams/{team_id}/members/bulk_update`. + +Every read runs on the writer inside the batch transaction, so the write plan can never be +built from a lagging read replica. Any budget row that more than one membership points at, +the team's shared default included, is cloned before it is written, so raising one member's +cap never moves another member's. +""" + +from collections.abc import Sequence +from datetime import datetime, timedelta +from types import MappingProxyType +from typing import TYPE_CHECKING, Final + +from pydantic import BaseModel, ConfigDict + +from litellm.litellm_core_utils.safe_json_dumps import safe_dumps +from litellm.proxy._types import ( + LiteLLM_TeamTable, + LitellmTableNames, + LitellmUserRoles, + Member, + UserAPIKeyAuth, +) +from litellm.proxy.auth.auth_checks import invalidate_team_member_spend_state +from litellm.proxy.common_utils.user_api_key_cache import UserApiKeyCache +from litellm.proxy.db.routing_prisma_wrapper import WriterPinnedClient +from litellm.proxy.management_endpoints.common_utils import ( + _is_user_org_admin_for_team, # pyright: ignore[reportPrivateUsage] # same check /team/member_update uses + _is_user_team_admin, # pyright: ignore[reportPrivateUsage] # same check /team/member_update uses + _upsert_budget_and_membership, # pyright: ignore[reportPrivateUsage] # the single-member write, shared so the two surfaces cannot drift + member_budget_patch, +) +from litellm.proxy.management_helpers.audit_logs import create_object_audit_log +from litellm.proxy.management_helpers.bulk_user_deletion import ( + _duplicate_member_indexes, # pyright: ignore[reportPrivateUsage] # same duplicate rule as members/bulk_delete + _eq_filter, # pyright: ignore[reportPrivateUsage] # same prisma filter shape as members/bulk_delete + _forbidden, # pyright: ignore[reportPrivateUsage] # same problem shape as members/bulk_delete + _in_filter, # pyright: ignore[reportPrivateUsage] # same prisma filter shape as members/bulk_delete + _team_not_found, # pyright: ignore[reportPrivateUsage] # same problem shape as members/bulk_delete + _team_users_filter, # pyright: ignore[reportPrivateUsage] # same prisma filter shape as members/bulk_delete +) +from litellm.proxy.utils import PrismaClient +from litellm.repositories.team_repository import TeamRepository +from litellm.types.proxy.management_endpoints.team_endpoints import ( + BulkTeamMemberBudgetUpdateRequest, + TeamMemberBudgetPatch, + TeamMemberBudgetUpdateResult, +) + +if TYPE_CHECKING: + from prisma import Prisma + from prisma import models as prisma_models + + from litellm.repositories.prisma_protocols import TableActions + +_BATCH_TX_TIMEOUT: Final = timedelta(seconds=60) +_NO_METADATA: Final = MappingProxyType({}) +_WITH_BUDGET: Final = MappingProxyType({"litellm_budget_table": True}) + + +def _membership_tx_db(tx: "Prisma") -> "TableActions[prisma_models.LiteLLM_TeamMembership]": + return tx.litellm_teammembership # pyright: ignore[reportReturnType] # TableActions widens the generated inputs to Mapping, as the repositories do + + +def _budget_tx_db(tx: "Prisma") -> "TableActions[prisma_models.LiteLLM_BudgetTable]": + return tx.litellm_budgettable # pyright: ignore[reportReturnType] # TableActions widens the generated inputs to Mapping, as the repositories do + + +def _roster_user_id(member: TeamMemberBudgetPatch, roster: Sequence[Member]) -> str | None: + """The team member this row addresses, or None when it names nobody on the team.""" + if member.user_id is not None: + return member.user_id if any(m.user_id == member.user_id for m in roster) else None + return next((m.user_id for m in roster if m.user_email is not None and m.user_email == member.user_email), None) + + +def _team_default_budget_id(team: LiteLLM_TeamTable) -> str | None: + raw: Final = (team.metadata or _NO_METADATA).get("team_member_budget_id") + return raw if isinstance(raw, str) else None + + +async def _shared_budget_ids(tx: "Prisma", budget_ids: frozenset[str]) -> frozenset[str]: + """The rows in ``budget_ids`` more than one membership points at, counted across every + team so a row shared with another team is protected too.""" + if not budget_ids: + return frozenset() + rows: Final = await _membership_tx_db(tx).find_many(where=_in_filter("budget_id", budget_ids)) + return frozenset(budget_id for budget_id in budget_ids if sum(1 for row in rows if row.budget_id == budget_id) > 1) + + +class _AuditedMemberBudget(BaseModel): + """One member's limits as the audit log's before/after values record them.""" + + model_config = ConfigDict(frozen=True) + + user_id: str + budget_id: str | None = None + max_budget: float | None = None + tpm_limit: int | None = None + rpm_limit: int | None = None + budget_duration: str | None = None + budget_reset_at: datetime | None = None + allowed_models: tuple[str, ...] | None = None + + +class _AuditedMemberBudgets(BaseModel): + """The audit-log columns hold a JSON object, so the per-member list is nested under a key.""" + + model_config = ConfigDict(frozen=True) + + team_member_budgets: tuple[_AuditedMemberBudget, ...] + + +def _audited_member_budget(row: "prisma_models.LiteLLM_TeamMembership") -> _AuditedMemberBudget: + budget: Final = row.litellm_budget_table + if budget is None: + return _AuditedMemberBudget(user_id=row.user_id, budget_id=row.budget_id) + return _AuditedMemberBudget( + user_id=row.user_id, + budget_id=row.budget_id, + max_budget=budget.max_budget, + tpm_limit=budget.tpm_limit, + rpm_limit=budget.rpm_limit, + budget_duration=budget.budget_duration, + budget_reset_at=budget.budget_reset_at, + allowed_models=tuple(budget.allowed_models), + ) + + +def _limits_audit_value(rows: "Sequence[prisma_models.LiteLLM_TeamMembership]") -> str: + """Serialize the members' limits for an audit-log value, dropping the limits they do not set.""" + return safe_dumps( + _AuditedMemberBudgets( + team_member_budgets=tuple(_audited_member_budget(row) for row in sorted(rows, key=lambda row: row.user_id)) + ).model_dump(exclude_none=True, mode="json") + ) + + +def _result( + member: TeamMemberBudgetPatch, + user_id: str | None, + error: str | None, + budget_of: "MappingProxyType[str, prisma_models.LiteLLM_BudgetTable | None]", + team_default_max_budget: float | None, +) -> TeamMemberBudgetUpdateResult: + if error is not None or user_id is None: + return TeamMemberBudgetUpdateResult( + user_id=member.user_id, + user_email=member.user_email, + success=False, + error=error or "User not found in team", + ) + budget: Final = budget_of.get(user_id) + own_max_budget: Final = budget.max_budget if budget is not None else None + inherits: Final = own_max_budget is None and team_default_max_budget is not None and team_default_max_budget > 0 + return TeamMemberBudgetUpdateResult( + user_id=user_id, + user_email=member.user_email, + success=True, + budget_id=budget.budget_id if budget is not None else None, + max_budget=team_default_max_budget if inherits else own_max_budget, + max_budget_source=("team_default" if inherits else "member" if own_max_budget is not None else None), + tpm_limit=budget.tpm_limit if budget is not None else None, + rpm_limit=budget.rpm_limit if budget is not None else None, + budget_duration=budget.budget_duration if budget is not None else None, + allowed_models=tuple(budget.allowed_models) if budget is not None else None, + ) + + +async def bulk_update_team_member_budgets( + team_id: str, + data: BulkTeamMemberBudgetUpdateRequest, + user_api_key_dict: UserAPIKeyAuth, + prisma_client: PrismaClient, + user_api_key_cache: UserApiKeyCache, + litellm_proxy_admin_name: str, + litellm_changed_by: str | None = None, +) -> tuple[TeamMemberBudgetUpdateResult, ...]: + """Apply one merge patch of per-member limits per requested member, in one transaction.""" + team: Final = await TeamRepository(WriterPinnedClient(prisma_client.db)).find_by_id(team_id) + if team is None: + raise _team_not_found(team_id) + + if ( + user_api_key_dict.user_role != LitellmUserRoles.PROXY_ADMIN.value + and not _is_user_team_admin(user_api_key_dict=user_api_key_dict, team_obj=team) + and not await _is_user_org_admin_for_team(user_api_key_dict=user_api_key_dict, team_obj=team) + ): + raise _forbidden( + "Call not allowed. User not proxy admin OR team admin OR org admin for this team. " + f"route='/management/v1/teams/{team_id}/members/bulk_update'" + ) + + roster: Final = team.members_with_roles or () + named: Final = tuple(_roster_user_id(member, roster) for member in data.members) + duplicates: Final = _duplicate_member_indexes(data.members) | frozenset( + index for index, user_id in enumerate(named) if user_id is not None and user_id in named[:index] + ) + applied: Final = tuple( + (index, user_id) for index, user_id in enumerate(named) if user_id is not None and index not in duplicates + ) + if not applied: + return tuple( + _result( + member, None, "Duplicate member in request" if index in duplicates else None, MappingProxyType({}), None + ) + for index, member in enumerate(data.members) + ) + + user_ids: Final = sorted(user_id for _, user_id in applied) + default_budget_id: Final = _team_default_budget_id(team) + team_members_filter: Final = _team_users_filter(team_id, user_ids) + + async with prisma_client.tx(timeout=_BATCH_TX_TIMEOUT) as tx: + memberships: Final = await _membership_tx_db(tx).find_many(where=team_members_filter, include=_WITH_BUDGET) + budget_id_of: Final = MappingProxyType({m.user_id: m.budget_id for m in memberships}) + shared: Final = await _shared_budget_ids( + tx, frozenset(budget_id for budget_id in budget_id_of.values() if budget_id is not None) + ) + for index, user_id in applied: + await _upsert_budget_and_membership( + tx=tx, + team_id=team_id, + user_id=user_id, + existing_budget_id=budget_id_of.get(user_id), + user_api_key_dict=user_api_key_dict, + budget_patch=member_budget_patch(data.members[index]), + team_default_budget_id=default_budget_id, + shared_budget_ids=shared, + ) + written: Final = await _membership_tx_db(tx).find_many(where=team_members_filter, include=_WITH_BUDGET) + team_default: Final = ( + await _budget_tx_db(tx).find_unique(where=_eq_filter("budget_id", default_budget_id)) + if default_budget_id is not None + else None + ) + + for user_id in user_ids: + await invalidate_team_member_spend_state( + user_id=user_id, team_id=team_id, user_api_key_cache=user_api_key_cache + ) + + await create_object_audit_log( + object_id=team_id, + action="updated", + litellm_changed_by=litellm_changed_by, + user_api_key_dict=user_api_key_dict, + litellm_proxy_admin_name=litellm_proxy_admin_name, + table_name=LitellmTableNames.TEAM_TABLE_NAME, + before_value=_limits_audit_value(memberships), + after_value=_limits_audit_value(written), + ) + + budget_of: Final = MappingProxyType({m.user_id: m.litellm_budget_table for m in written}) + return tuple( + _result( + member, + named[index], + "Duplicate member in request" if index in duplicates else None, + budget_of, + team_default.max_budget if team_default is not None else None, + ) + for index, member in enumerate(data.members) + ) diff --git a/litellm/proxy/pass_through_endpoints/llm_passthrough_endpoints.py b/litellm/proxy/pass_through_endpoints/llm_passthrough_endpoints.py index b9b8cb3a22b..f251b3b052c 100644 --- a/litellm/proxy/pass_through_endpoints/llm_passthrough_endpoints.py +++ b/litellm/proxy/pass_through_endpoints/llm_passthrough_endpoints.py @@ -525,6 +525,42 @@ async def mistral_proxy_route( return received_value +@router.api_route( + "/typesafe/{endpoint:path}", + methods=["GET", "POST"], # mutable-ok: FastAPI route metadata requires a list + tags=["TypeSafe AI Pass-through", "pass-through"], # mutable-ok: FastAPI route metadata requires a list +) +async def typesafe_proxy_route( + endpoint: str, + request: Request, + fastapi_response: Response, + user_api_key_dict: Annotated[UserAPIKeyAuth, Depends(user_api_key_auth)], +): + """[Docs](https://docs.litellm.ai/docs/pass_through/typesafe)""" + base_target_url: Final = get_secret_str("TYPESAFE_API_BASE") or "https://api.typesafe.ai" + encoded_endpoint: Final = httpx.URL(endpoint).path + normalized_endpoint: Final = encoded_endpoint if encoded_endpoint.startswith("/") else f"/{encoded_endpoint}" + base_url: Final = httpx.URL(base_target_url) + updated_url: Final = base_url.copy_with( + path=HttpPassThroughEndpointHelpers.join_base_and_endpoint_path(base_url, normalized_endpoint), + ) + typesafe_api_key: Final = passthrough_endpoint_router.get_credentials( + custom_llm_provider="typesafe", + region_name=None, + ) + endpoint_func: Final = create_pass_through_route( + endpoint=endpoint, + target=str(updated_url), + custom_headers={ # mutable-ok: pass-through request headers require a mutable mapping + "Authorization": f"Bearer {typesafe_api_key}", + "Content-Type": "application/json", + }, + custom_llm_provider="typesafe", + is_streaming_request=False, + ) + return await endpoint_func(request, fastapi_response, user_api_key_dict) + + @router.api_route( "/milvus/{endpoint:path}", methods=["GET", "POST", "PUT", "DELETE", "PATCH"], diff --git a/litellm/proxy/pass_through_endpoints/llm_provider_handlers/typesafe_passthrough_logging_handler.py b/litellm/proxy/pass_through_endpoints/llm_provider_handlers/typesafe_passthrough_logging_handler.py new file mode 100644 index 00000000000..9b196660c2c --- /dev/null +++ b/litellm/proxy/pass_through_endpoints/llm_provider_handlers/typesafe_passthrough_logging_handler.py @@ -0,0 +1,117 @@ +from collections.abc import Mapping +from datetime import datetime +from typing import Final + +import httpx +from pydantic import BaseModel, TypeAdapter, ValidationError + +import litellm +from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj +from litellm.litellm_core_utils.litellm_logging import ( + get_standard_logging_object_payload, # pyright: ignore[reportUnknownVariableType] # legacy helper has an untyped signature +) +from litellm.proxy._types import PassThroughEndpointLoggingTypedDict +from litellm.types.utils import ModelResponse, StandardPassThroughResponseObject, Usage + + +class _TypeSafeUsage(BaseModel): + input_tokens: int = 0 + output_tokens: int = 0 + + +class _TypeSafeResponse(BaseModel): + model: str | None = None + usage: _TypeSafeUsage | None = None + + +class _RegistryPricing(BaseModel): + input_cost_per_token: float = 0.0 + output_cost_per_token: float = 0.0 + + +_TYPESAFE_RESPONSE_ADAPTER: Final = TypeAdapter(_TypeSafeResponse) +_REGISTRY_PRICING_ADAPTER: Final = TypeAdapter(_RegistryPricing) + + +def _parse_typesafe_response(response_body: Mapping[str, object]) -> _TypeSafeResponse: + try: + return _TYPESAFE_RESPONSE_ADAPTER.validate_python(response_body) + except ValidationError: + return _TypeSafeResponse() + + +def _pricing_for(model_keys: tuple[str, ...]) -> _RegistryPricing: + for model_key in model_keys: + if model_key not in litellm.model_cost: # pyright: ignore[reportUnknownMemberType] # registry is dynamically typed + continue + try: + return _REGISTRY_PRICING_ADAPTER.validate_python( + litellm.model_cost[model_key] # pyright: ignore[reportUnknownMemberType] # registry is dynamically typed + ) + except ValidationError: + continue + return _RegistryPricing() + + +class TypeSafePassthroughLoggingHandler: + @staticmethod + def typesafe_passthrough_handler( + httpx_response: httpx.Response, + response_body: Mapping[str, object], + logging_obj: LiteLLMLoggingObj, + url_route: str, + result: str, + start_time: datetime, + end_time: datetime, + cache_hit: bool, + request_body: Mapping[str, object], + **kwargs: object, + ) -> PassThroughEndpointLoggingTypedDict: + response: Final = _parse_typesafe_response(response_body) + response_model: Final = response.model + request_model_value: Final = request_body.get("model") + request_model: Final = request_model_value if isinstance(request_model_value, str) else None + logged_model: Final = response_model or request_model or "unknown" + model_name: Final = f"typesafe/{logged_model}" + usage: Final = response.usage or _TypeSafeUsage() + input_tokens: Final = usage.input_tokens + output_tokens: Final = usage.output_tokens + candidate_model_keys: Final = tuple( + f"typesafe/{model}" for model in (response_model, request_model) if model is not None + ) + pricing: Final = _pricing_for(candidate_model_keys) + response_cost: Final = ( + input_tokens * pricing.input_cost_per_token + output_tokens * pricing.output_cost_per_token + ) + usage_object: Final = Usage( + prompt_tokens=input_tokens, + completion_tokens=output_tokens, + total_tokens=input_tokens + output_tokens, + ) + updated_kwargs: Final = { # mutable-ok: pass-through logging contract requires mutable kwargs + **kwargs, + "model": model_name, + "custom_llm_provider": "typesafe", + "response_cost": response_cost, + "combined_usage_object": usage_object, + } + logging_obj.model_call_details.update( + model=model_name, + custom_llm_provider="typesafe", + response_cost=response_cost, + ) + standard_logging_object: Final = get_standard_logging_object_payload( + kwargs=updated_kwargs, + init_response_obj=ModelResponse(model=model_name, usage=usage_object), + start_time=start_time, + end_time=end_time, + logging_obj=logging_obj, + status="success", + ) + return { # mutable-ok: pass-through logging contract requires mutable result + "result": StandardPassThroughResponseObject(response=result), + "kwargs": { # mutable-ok: pass-through logging contract requires mutable kwargs + **updated_kwargs, + "standard_logging_object": standard_logging_object, + }, + } diff --git a/litellm/proxy/pass_through_endpoints/success_handler.py b/litellm/proxy/pass_through_endpoints/success_handler.py index 76a471302f4..699caae819d 100644 --- a/litellm/proxy/pass_through_endpoints/success_handler.py +++ b/litellm/proxy/pass_through_endpoints/success_handler.py @@ -1,5 +1,6 @@ import json from datetime import datetime +from types import MappingProxyType from typing import Any, Final from urllib.parse import urlparse @@ -256,6 +257,25 @@ class PassThroughEndpointLogging: ) standard_logging_response_object = comprehend_medical_handler_result["result"] # rebind-ok: elif-chain kwargs = comprehend_medical_handler_result["kwargs"] # rebind-ok: elif-chain contract + elif self.is_typesafe_route(custom_llm_provider): + from .llm_provider_handlers.typesafe_passthrough_logging_handler import ( + TypeSafePassthroughLoggingHandler, + ) + + typesafe_handler_result: Final = TypeSafePassthroughLoggingHandler.typesafe_passthrough_handler( + httpx_response=httpx_response, + response_body=response_body if isinstance(response_body, dict) else MappingProxyType({}), + logging_obj=logging_obj, + url_route=url_route, + result=result, + start_time=start_time, + end_time=end_time, + cache_hit=cache_hit, + request_body=request_body, + **kwargs, + ) + standard_logging_response_object = typesafe_handler_result["result"] + kwargs = typesafe_handler_result["kwargs"] elif self.is_vertex_ai_live_route(url_route): from .llm_provider_handlers.vertex_ai_live_passthrough_logging_handler import ( VertexAILivePassthroughLoggingHandler, @@ -389,6 +409,9 @@ class PassThroughEndpointLogging: def is_comprehend_medical_route(self, custom_llm_provider: str | None) -> bool: return custom_llm_provider == "comprehendmedical" + def is_typesafe_route(self, custom_llm_provider: str | None) -> bool: + return custom_llm_provider == "typesafe" + def is_langfuse_route(self, url_route: str): parsed_url: Final = urlparse(url_route) for route in self.TRACKED_LANGFUSE_ROUTES: diff --git a/litellm/proxy/spend_tracking/spend_tracking_utils.py b/litellm/proxy/spend_tracking/spend_tracking_utils.py index 09d719202ca..5a3a3f6c2f4 100644 --- a/litellm/proxy/spend_tracking/spend_tracking_utils.py +++ b/litellm/proxy/spend_tracking/spend_tracking_utils.py @@ -39,6 +39,7 @@ from litellm.litellm_core_utils.litellm_logging import ( is_valid_sha256_hash, request_model_access_groups_from_litellm_params, ) +from litellm.litellm_core_utils.ptu_pricing import azure_spillover from litellm.litellm_core_utils.safe_json_dumps import safe_dumps, strip_null_bytes from litellm.proxy._types import SpendLogsMetadata, SpendLogsPayload, SpendLogsRouterMetadata from litellm.proxy.route_llm_request import ProxyModelNotFoundError @@ -47,6 +48,7 @@ from litellm.proxy.utils import PrismaClient, hash_token from litellm.types.router import DeploymentTypedDict, LiteLLM_Params from litellm.types.utils import ( PROMPT_CARRYING_GUARDRAIL_FIELDS, + AzureSpillover, CallTypes, CostBreakdown, LlmProviders, @@ -133,6 +135,9 @@ def _get_router_metadata_for_spend_log( ) +_STAMPED_METADATA_KEYS: Final = frozenset(("router_metadata", "azure_spillover")) + + def _get_spend_logs_metadata( metadata: dict | None, applied_guardrails: list[str] | None = None, @@ -150,6 +155,7 @@ def _get_spend_logs_metadata( litellm_call_id: str | None = None, autorouter_savings: float | None = None, router_metadata: SpendLogsRouterMetadata | None = None, + azure_spillover: AzureSpillover | None = None, ) -> SpendLogsMetadata: if metadata is None: return SpendLogsMetadata( @@ -191,6 +197,7 @@ def _get_spend_logs_metadata( litellm_gateway_injected_cache=None, litellm_call_id=litellm_call_id, router_metadata=router_metadata, + azure_spillover=azure_spillover, ) verbose_proxy_logger.debug( "getting payload for SpendLogs, available keys in metadata: " + str(list(metadata.keys())) @@ -198,8 +205,9 @@ def _get_spend_logs_metadata( # Filter the metadata dictionary to include only the specified keys clean_metadata: Final = SpendLogsMetadata( - **{key: metadata.get(key) for key in SpendLogsMetadata.__annotations__ if key != "router_metadata"}, + **{key: metadata.get(key) for key in SpendLogsMetadata.__annotations__ if key not in _STAMPED_METADATA_KEYS}, router_metadata=router_metadata, + azure_spillover=azure_spillover, ) _raw_key: Final = clean_metadata.get("user_api_key") _trusted_hash: Final = metadata.get("user_api_key_hash") @@ -573,6 +581,15 @@ def get_logging_payload( selected_provider=custom_llm_provider, router_correlation_id=litellm_call_id, ), + azure_spillover=azure_spillover( + response_headers=kwargs.get("response_headers") + if isinstance(kwargs.get("response_headers"), Mapping) + else None, + additional_headers=standard_logging_payload["hidden_params"].get("additional_headers") + if standard_logging_payload is not None + and isinstance(standard_logging_payload.get("hidden_params"), Mapping) + else None, + ), ) special_usage_fields: Final = ["completion_tokens", "prompt_tokens", "total_tokens"] diff --git a/litellm/types/proxy/management_endpoints/team_endpoints.py b/litellm/types/proxy/management_endpoints/team_endpoints.py index 5f5be81ee4b..4524c47ec38 100644 --- a/litellm/types/proxy/management_endpoints/team_endpoints.py +++ b/litellm/types/proxy/management_endpoints/team_endpoints.py @@ -1,6 +1,6 @@ from typing import Any, Final, Literal -from pydantic import BaseModel, ConfigDict, Field, model_validator +from pydantic import BaseModel, ConfigDict, Field, field_validator, model_validator from litellm.proxy._types import ( KeyManagementRoutes, @@ -10,12 +10,15 @@ from litellm.proxy._types import ( Member, MemberDeleteRequest, ) +from litellm.proxy.common_utils.timezone_utils import budget_duration_error from litellm.types.proxy.management_endpoints.management_v1 import ResourceResponse TeamIdSearchMatch = Literal["exact", "prefix"] MAX_BULK_TEAM_MEMBER_DELETES: Final = 500 +MAX_BULK_TEAM_MEMBER_BUDGET_UPDATES: Final = 500 + class GetTeamMemberPermissionsRequest(BaseModel): """Request to get the team member permissions for a team""" @@ -123,7 +126,7 @@ class BulkTeamMemberAddResponse(BaseModel): class TeamMemberRef(MemberDeleteRequest): - """One member to remove, named by exactly one of `user_id` or `user_email`.""" + """One member, named by exactly one of `user_id` or `user_email`.""" model_config = ConfigDict(extra="forbid") @@ -155,6 +158,55 @@ class BulkTeamMemberDeleteResponse(ResourceResponse[tuple[TeamMemberDeleteResult """`{data: [...]}` with one `TeamMemberDeleteResult` per requested member, in request order.""" +class TeamMemberBudgetPatch(TeamMemberRef): + """One member's per-member limits, merge-patch style: a field left out of the row is + untouched, a field sent as null is cleared, and clearing the last limit drops the + member back to the team default.""" + + max_budget_in_team: float | None = None + tpm_limit: int | None = None + rpm_limit: int | None = None + budget_duration: str | None = None + allowed_models: tuple[str, ...] | None = None + + @field_validator("budget_duration") + @classmethod + def persistable_budget_duration(cls, value: str | None) -> str | None: + error: Final = budget_duration_error(value) + if error is not None: + raise ValueError(error) + return value + + +class BulkTeamMemberBudgetUpdateRequest(BaseModel): + """Body of `POST /management/v1/teams/{team_id}/members/bulk_update`.""" + + model_config = ConfigDict(extra="forbid") + + members: tuple[TeamMemberBudgetPatch, ...] = Field(min_length=1, max_length=MAX_BULK_TEAM_MEMBER_BUDGET_UPDATES) + + +class TeamMemberBudgetUpdateResult(BaseModel): + """Outcome for one requested member, in request order, carrying the limits in force + after the write rather than the ones that were asked for.""" + + user_id: str | None = None + user_email: str | None = None + success: bool + error: str | None = None + budget_id: str | None = None + max_budget: float | None = None + max_budget_source: Literal["member", "team_default"] | None = None + tpm_limit: int | None = None + rpm_limit: int | None = None + budget_duration: str | None = None + allowed_models: tuple[str, ...] | None = None + + +class BulkTeamMemberBudgetUpdateResponse(ResourceResponse[tuple[TeamMemberBudgetUpdateResult, ...]]): + """`{data: [...]}` with one `TeamMemberBudgetUpdateResult` per requested member, in request order.""" + + class TeamMemberInfoResponse(LiteLLM_TeamMembership): """Response for GET /team/{team_id}/members/me ā€” caller's own membership row.""" diff --git a/litellm/types/utils.py b/litellm/types/utils.py index aaa16fd2d44..9f6fcf834c5 100644 --- a/litellm/types/utils.py +++ b/litellm/types/utils.py @@ -348,6 +348,7 @@ class ModelInfoBase(ProviderSpecificModelInfo, total=False): "audio_transcription", "audio_speech", "responses", + "evaluation", "ocr", "realtime", ] @@ -3075,6 +3076,12 @@ class StandardLoggingMetadata(StandardLoggingUserAPIKeyMetadata): team_id: str | None +class AzureSpillover(TypedDict): + """Spillover Azure reports in its response headers for a request it served from pay-as-you-go capacity.""" + + from_deployment: ReadOnly[str | None] + + class StandardLoggingAdditionalHeaders(TypedDict, total=False): x_ratelimit_limit_requests: int x_ratelimit_limit_tokens: int @@ -3754,6 +3761,8 @@ agentic_loop_internal_litellm_params: Final = [ # the provider. TRUSTED_CALLBACK_VARS_FIELD: Final = "litellm_trusted_callback_vars" +ADDRESSED_RESPONSE_ID_FIELD: Final = "_litellm_addressed_response_id" + # Bedrock managed-batch deployment config, read from litellm_params by the batch and # files transformations. Listed for the same reason as the fields above: these sit on # a deployment that also serves chat, so leaking them into extra_body makes Bedrock @@ -3768,7 +3777,7 @@ bedrock_batch_litellm_params: Final = ( all_litellm_params = ( agentic_loop_internal_litellm_params - + [TRUSTED_CALLBACK_VARS_FIELD, *bedrock_batch_litellm_params] + + [TRUSTED_CALLBACK_VARS_FIELD, ADDRESSED_RESPONSE_ID_FIELD, *bedrock_batch_litellm_params] + [ "metadata", "litellm_metadata", diff --git a/model_prices_and_context_window.json b/model_prices_and_context_window.json index 96cdbcffd90..602c3ec7037 100644 --- a/model_prices_and_context_window.json +++ b/model_prices_and_context_window.json @@ -23788,7 +23788,7 @@ "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": false + "supports_vision": true }, "fireworks_ai/accounts/fireworks/models/mixtral-8x22b-instruct-hf": { "input_cost_per_token": 1.2e-06, @@ -24114,7 +24114,7 @@ "supports_reasoning": true, "supports_response_schema": true, "supports_tool_choice": true, - "supports_vision": false + "supports_vision": true }, "fireworks_ai/qwen3p7-plus": { "cache_read_input_token_cost": 8e-08, @@ -69184,6 +69184,27 @@ "supports_reasoning": true, "supports_vision": true }, + "typesafe/jev-1.13.0": { + "input_cost_per_token": 4.2e-08, + "litellm_provider": "typesafe", + "mode": "evaluation", + "output_cost_per_token": 0.0, + "source": "https://docs.typesafe.ai/models" + }, + "typesafe/jev-latest": { + "input_cost_per_token": 4.2e-08, + "litellm_provider": "typesafe", + "mode": "evaluation", + "output_cost_per_token": 0.0, + "source": "https://docs.typesafe.ai/models" + }, + "typesafe/jev-preview": { + "input_cost_per_token": 4.2e-08, + "litellm_provider": "typesafe", + "mode": "evaluation", + "output_cost_per_token": 0.0, + "source": "https://docs.typesafe.ai/models" + }, "wandb/zai-org/GLM-5.3-Flash": { "cache_read_input_token_cost": 5e-08, "input_cost_per_token": 1.5e-07, diff --git a/model_prices_and_context_window.schema.json b/model_prices_and_context_window.schema.json index 130cc6873fa..f924df1f1b2 100644 --- a/model_prices_and_context_window.schema.json +++ b/model_prices_and_context_window.schema.json @@ -427,6 +427,7 @@ "chat", "completion", "embedding", + "evaluation", "guardrail", "image_edit", "image_generation", diff --git a/pyproject.toml b/pyproject.toml index 615f4b8d0ab..dfe84a28d52 100644 --- a/pyproject.toml +++ b/pyproject.toml @@ -174,7 +174,7 @@ proxy-runtime = [ [project.scripts] litellm = "litellm:run_server" lite = "litellm.proxy.client.cli:cli" -litellm-proxy = "litellm.proxy.client.cli:cli" +litellm-proxy = "litellm.proxy.client.cli:litellm_proxy_cli" [dependency-groups] dev = [ diff --git a/tests/e2e/batches/COVERAGE.md b/tests/e2e/batches/COVERAGE.md index 919c39f21a2..b36d8937ad0 100644 --- a/tests/e2e/batches/COVERAGE.md +++ b/tests/e2e/batches/COVERAGE.md @@ -20,6 +20,7 @@ failures are hard test failures (see `tests/e2e/CLAUDE.md`). | Azure | yes | yes | yes | yes | yes (byte-verbatim) | Azure Files | | Vertex AI | yes | yes | yes | yes | yes (provider-transformed) | GCS (`gcs_bucket_name` / `GCS_BUCKET_NAME` on model) | | Bedrock | yes (unified only) | yes | yes | yes (unfiltered managed list) | yes (provider-transformed) | S3 (`s3_bucket_name` + `aws_*` + `AWS_BATCH_ROLE_ARN` on model) | +| Bedrock GovCloud (`us-gov-west-1`) | yes (unified only) | yes | no | no | yes (provider-transformed) | S3 (`s3_bucket_name` + `aws_*` on model, resolved from `AWS_GOVCLOUD_ACCESS_KEY_ID` / `AWS_GOVCLOUD_SECRET_ACCESS_KEY` / `AWS_GOVCLOUD_BATCH_S3_BUCKET` / `AWS_GOVCLOUD_BATCH_ROLE_ARN`) | Bedrock cancel maps to `StopModelInvocationJob` and comes back `cancelling`; the lifecycle asserts it the same way it does for OpenAI (`_CANCEL_ASSERTED_PROVIDERS`). diff --git a/tests/e2e/batches/test_batches_e2e.py b/tests/e2e/batches/test_batches_e2e.py index c4b699190b8..eff8f297f25 100644 --- a/tests/e2e/batches/test_batches_e2e.py +++ b/tests/e2e/batches/test_batches_e2e.py @@ -21,21 +21,18 @@ import os import re import time from datetime import datetime, timedelta, timezone +from typing import Final import pytest -from pydantic import BaseModel - -from e2e_config import MASTER_KEY, PROXY_BASE_URL, unique_marker - from batch_cleanup import cleanup_batch, cleanup_file from batch_client import ( AZURE_FILE_EXPIRY_SECONDS, - batch_upload_form, UPLOAD_FILENAME, BatchClient, BatchCreateBody, BatchObject, FileObject, + batch_upload_form, is_model_access_denied, is_result_access_denied, ) @@ -57,6 +54,7 @@ from capabilities import ( openai_batch_params, raw_id_matches_provider, ) +from e2e_config import MASTER_KEY, PROXY_BASE_URL, unique_marker from e2e_http import ( FileUploadForm, Result, @@ -68,6 +66,7 @@ from e2e_http import ( ) from lifecycle import ResourceManager from models import KeyGenerateBody, KeyMetadata, LiteLLMParamsBody, SpendLogRow +from pydantic import BaseModel, Field pytestmark = pytest.mark.e2e @@ -75,6 +74,25 @@ CREATED_BATCH_STATUSES = {"validating", "in_progress", "finalizing"} BATCH_CANCEL_DELAY_SECONDS = 2 BATCH_TERMINAL_BEFORE_CANCEL = {"failed", "cancelled", "expired"} BATCH_OP_RETRIES = 5 + + +class _GovCloudBedrockContent(BaseModel): + text: str + + +class _GovCloudBedrockMessage(BaseModel): + content: tuple[_GovCloudBedrockContent, ...] + + +class _GovCloudBedrockInput(BaseModel): + messages: tuple[_GovCloudBedrockMessage, ...] + + +class _GovCloudBedrockRecord(BaseModel): + record_id: str = Field(alias="recordId") + model_input: _GovCloudBedrockInput = Field(alias="modelInput") + + # Azure / Vertex cancel and the pre-cancel re-retrieve are provider-side flakes # (connection refused, brief 500s) and the registry only has one basic cell per # provider (shared across scenarios). Create + retrieve already prove routing; @@ -1006,6 +1024,91 @@ class TestBedrockBatchAssumeRole: assert fetched.id == batch.id +GOVCLOUD_REGION: Final = "us-gov-west-1" +GOVCLOUD_RAW_MODEL: Final = "bedrock/amazon.nova-lite-v1:0" + + +def _govcloud_params() -> LiteLLMParamsBody: + return LiteLLMParamsBody( + model=GOVCLOUD_RAW_MODEL, + aws_access_key_id="os.environ/AWS_GOVCLOUD_ACCESS_KEY_ID", + aws_secret_access_key="os.environ/AWS_GOVCLOUD_SECRET_ACCESS_KEY", + aws_region_name=GOVCLOUD_REGION, + s3_region_name=GOVCLOUD_REGION, + s3_bucket_name="os.environ/AWS_GOVCLOUD_BATCH_S3_BUCKET", + s3_access_key_id="os.environ/AWS_GOVCLOUD_ACCESS_KEY_ID", + s3_secret_access_key="os.environ/AWS_GOVCLOUD_SECRET_ACCESS_KEY", + aws_batch_role_arn="os.environ/AWS_GOVCLOUD_BATCH_ROLE_ARN", + ) + + +class TestBedrockBatchGovCloud: + """Bedrock batch lifecycle in the AWS GovCloud partition (us-gov-west-1). + + The deployment carries a GovCloud region for both Bedrock and S3, so the proxy has to + sign the file upload against the us-gov S3 endpoint and submit the job to the us-gov + Bedrock endpoint. Commercial-partition hostnames or arn:aws: ARNs reject the GovCloud + key, so a partition regression fails the upload instead of passing silently. + """ + + @pytest.mark.covers( + "llm.batches.bedrock.govcloud_partition.nonstream.works", + "llm.files.bedrock.govcloud_partition.nonstream.works", + exercised_on=["batches", "files"], + ) + def test_unified_file_upload_and_batch_create_in_govcloud( + self, client: BatchClient, resources: ResourceManager + ) -> None: + model_name: Final = batch_model_name("bedrock-govcloud-batch") + model_id: Final = client.create_model(model_name, _govcloud_params()) + resources.defer(lambda: client.delete_model(model_id)) + key: Final = resources.key() + file: Final = unwrap( + client.upload_file( + content=render_jsonl(GOVCLOUD_RAW_MODEL), + form=FileUploadForm(purpose="batch", target_model_names=model_name), + key=key, + ) + ) + resources.defer(lambda: cleanup_file(client, file.id, key=key)) + assert_file_object(file, provider="bedrock") + + downloaded: Final = client.proxy.transport.download( + f"/v1/files/{file.id}/content", + headers=client.proxy.transport.bearer(key), + ) + assert downloaded.status_code == 200, ( + f"GovCloud file content must be 200, got {downloaded.status_code}: {downloaded.body[:300]}" + ) + downloaded_lines: Final = downloaded.body.strip().splitlines() + assert len(downloaded_lines) == 1, ( + f"GovCloud file content download must contain one JSONL record, got {len(downloaded_lines)}" + ) + downloaded_record: Final = _GovCloudBedrockRecord.model_validate(json.loads(downloaded_lines[0])) + assert downloaded_record.record_id == "req-1", ( + f"GovCloud file content must preserve the uploaded custom_id, got {downloaded_record.record_id!r}" + ) + assert downloaded_record.model_input.messages[0].content[0].text == "ping", ( + "GovCloud file content must preserve the uploaded message text" + ) + + created: Final = client.create_batch(body=BatchCreateBody(input_file_id=file.id), key=key) + require_successful_call(created) + batch: Final = BatchObject.model_validate_json(created.body) + resources.defer(lambda: cleanup_batch(client, batch.id, key=key)) + + assert is_managed_id(batch.id), ( + f"GovCloud create via target_model_names must return a managed batch id, got {batch.id!r}" + ) + assert batch.status in CREATED_BATCH_STATUSES, ( + f"GovCloud batch has non-transitional status {batch.status!r}" + ) + assert_batch_object(batch) + + fetched: Final = unwrap(client.retrieve_batch(batch.id, key=key)) + assert fetched.id == batch.id + + GEMINI_FILES_RAW_MODEL = "gemini-2.5-flash" diff --git a/tests/e2e/coverage_registry/llm_nonconversational.yaml b/tests/e2e/coverage_registry/llm_nonconversational.yaml index 635ea3f7ea5..50f9b9808b2 100644 --- a/tests/e2e/coverage_registry/llm_nonconversational.yaml +++ b/tests/e2e/coverage_registry/llm_nonconversational.yaml @@ -23,6 +23,7 @@ - {id: llm.batches.vertex.basic.nonstream.works, module: llm, tier: P0, subject_endpoint: batches, route: vertex, capability: basic, streaming: nonstream, assertions: [works], source: "batches/capabilities.py:98", rationale: "Vertex batches"} - {id: llm.batches.bedrock.basic.nonstream.works, module: llm, tier: P0, subject_endpoint: batches, route: bedrock_converse, capability: basic, streaming: nonstream, assertions: [works], source: "batches/capabilities.py:98", rationale: "Bedrock batches (encoded/unified only)"} - {id: llm.batches.bedrock.assume_role.nonstream.works, module: llm, tier: P0, subject_endpoint: batches, route: bedrock_converse, capability: assume_role, streaming: nonstream, assertions: [works], source: "test_batches_e2e.py", rationale: "Bedrock batch create under STS assume-role credentials"} +- {id: llm.batches.bedrock.govcloud_partition.nonstream.works, module: llm, tier: P0, subject_endpoint: batches, route: bedrock_converse, capability: govcloud_partition, streaming: nonstream, assertions: [works], source: "test_batches_e2e.py", rationale: "Bedrock batch create in the us-gov-west-1 partition"} - {id: llm.batches.bedrock.cancel.nonstream.works, module: llm, tier: P0, subject_endpoint: batches, route: bedrock_converse, capability: basic, streaming: nonstream, assertions: [works], source: "test_batches_e2e.py", rationale: "Bedrock batch cancel (StopModelInvocationJob) returns the same id with a cancelling/cancelled status"} - {id: llm.batches.bedrock.list.nonstream.works, module: llm, tier: P0, subject_endpoint: batches, route: bedrock_converse, capability: basic, streaming: nonstream, assertions: [works], source: "test_batches_e2e.py", rationale: "A Bedrock managed batch is present in the GET /v1/batches list envelope"} - {id: llm.batches.hosted_vllm.basic.nonstream.works, module: llm, tier: P1, subject_endpoint: batches, route: hosted_vllm, capability: basic, streaming: nonstream, assertions: [works], source: "test_batches_e2e.py", rationale: "hosted_vllm OpenAI-compatible batch create"} @@ -45,6 +46,7 @@ - {id: llm.files.azure_openai.upload.nonstream.works, module: llm, tier: P0, subject_endpoint: files, route: azure_openai, capability: basic, streaming: nonstream, assertions: [works], source: "batches/capabilities.py:45", rationale: "Azure file upload managed backend"} - {id: llm.files.vertex.upload.nonstream.works, module: llm, tier: P0, subject_endpoint: files, route: vertex, capability: basic, streaming: nonstream, assertions: [works], source: "batches/capabilities.py:52", rationale: "Vertex file upload to GCS"} - {id: llm.files.bedrock.upload.nonstream.works, module: llm, tier: P0, subject_endpoint: files, route: bedrock_converse, capability: basic, streaming: nonstream, assertions: [works], source: "batches/capabilities.py:59", rationale: "Bedrock file upload to S3"} +- {id: llm.files.bedrock.govcloud_partition.nonstream.works, module: llm, tier: P0, subject_endpoint: files, route: bedrock_converse, capability: govcloud_partition, streaming: nonstream, assertions: [works], source: "test_batches_e2e.py", rationale: "Bedrock file upload to an S3 bucket in the us-gov-west-1 partition"} - {id: llm.files.gemini.upload.nonstream.works, module: llm, tier: P1, subject_endpoint: files, route: gemini, capability: basic, streaming: nonstream, assertions: [works], source: "test_batches_e2e.py", rationale: "Gemini Files API upload via proxy"} - {id: llm.files.hosted_vllm.upload.nonstream.works, module: llm, tier: P1, subject_endpoint: files, route: hosted_vllm, capability: basic, streaming: nonstream, assertions: [works], source: "test_batches_e2e.py", rationale: "hosted_vllm OpenAI-compatible file upload"} - {id: llm.files.openai.require_managed_files_upload.nonstream.works, module: llm, tier: P1, subject_endpoint: files, route: openai, capability: input_validation, streaming: nonstream, assertions: [works], source: "test_managed_files_enforcement_e2e.py / LIT-5902", rationale: "With require_managed_files enabled, an upload without target_model_names and an upload carrying a model param are both rejected 400; runs only in the sequential managed-files stack phase (E2E_MANAGED_FILES_STACK)"} diff --git a/tests/e2e/coverage_registry/schema.py b/tests/e2e/coverage_registry/schema.py index 03d15f532b8..fa6dad90126 100644 --- a/tests/e2e/coverage_registry/schema.py +++ b/tests/e2e/coverage_registry/schema.py @@ -64,6 +64,7 @@ LlmCapability = Literal[ "assume_role", "basic", "count_tokens", + "govcloud_partition", "input_validation", "long_context_1m", "mid_conversation_system", diff --git a/tests/otel_tests/test_e2e_budgeting.py b/tests/otel_tests/test_e2e_budgeting.py index 44542558002..ca5058818e4 100644 --- a/tests/otel_tests/test_e2e_budgeting.py +++ b/tests/otel_tests/test_e2e_budgeting.py @@ -367,7 +367,7 @@ async def obtain_cli_sso_token_via_poll_flow( models: list[str], ) -> str: """ - Obtain a CLI SSO JWT through the same HTTP flow as `litellm-proxy login`: + Obtain a CLI SSO JWT through the same HTTP flow as `lite login`: /sso/cli/start -> (SSO callback) -> /sso/cli/complete -> /sso/cli/poll. When the proxy SSO session cache is not shared with the test runner (otel CI @@ -551,7 +551,7 @@ async def test_team_budget_enforcement(): @pytest.mark.asyncio async def test_team_budget_enforcement_cli_sso_token(): """ - Team budget enforcement for CLI SSO session tokens (litellm-proxy login JWT). + Team budget enforcement for CLI SSO session tokens (lite login JWT). 1. Create team with a tiny max_budget and a user on that team 2. Obtain a CLI SSO JWT (HTTP poll flow when Redis is shared, else mint) diff --git a/tests/test_litellm/integrations/test_prometheus_metric_name_consistency.py b/tests/test_litellm/integrations/test_prometheus_metric_name_consistency.py index 0932925d810..d648afcd087 100644 --- a/tests/test_litellm/integrations/test_prometheus_metric_name_consistency.py +++ b/tests/test_litellm/integrations/test_prometheus_metric_name_consistency.py @@ -8,9 +8,184 @@ configuration works correctly. Related issue: https://github.com/BerriAI/litellm/issues/18221 """ -from typing import get_args +import json +import re +from collections.abc import Iterator +from contextlib import contextmanager +from pathlib import Path +from types import MappingProxyType +from typing import Final, get_args import pytest +from prometheus_client import REGISTRY, Gauge +from prometheus_client.registry import Collector + +import litellm +from litellm.caching.redis_cache import _breaker_metrics +from litellm.integrations.prometheus import PrometheusLogger +from litellm.integrations.prometheus_services import PrometheusServicesLogger +from litellm.proxy.db.db_transaction_queue.spend_log_cleanup_metrics import SpendLogCleanupMetrics +from litellm.proxy.middleware.admission_control_middleware import create_prometheus_admission_metrics +from litellm.proxy.middleware.in_flight_requests_middleware import InFlightRequestsMiddleware + +_GRAFANA_DIR: Final = Path(__file__).parents[3] / "cookbook" / "litellm_proxy_server" / "grafana_dashboard" +_ALL_METRICS_DASHBOARD: Final = _GRAFANA_DIR / "dashboard_all_metrics" / "grafana_dashboard.json" +_LITELLM_DASHBOARDS: Final = (_ALL_METRICS_DASHBOARD, _GRAFANA_DIR / "dashboard_v2" / "grafana_dashboard.json") +_METRIC_TOKEN_RE: Final = re.compile(r"\blitellm_[a-z0-9_]+") +_BY_CLAUSE_RE: Final = re.compile(r"\bby\s*\([^)]*\)") +_EXPOSITION_SUFFIXES: Final = ("", "_total", "_bucket", "_sum", "_count", "_created") + + +def _registered_collectors() -> MappingProxyType[Collector, tuple[str, ...]]: + return MappingProxyType({collector: tuple(names) for collector, names in REGISTRY._collector_to_names.items()}) + + +def _unregister_everything() -> None: + for collector in tuple(REGISTRY._collector_to_names): + REGISTRY.unregister(collector) + + +def _register_if_absent(collectors: tuple[Collector, ...]) -> None: + for collector in collectors: + if collector not in REGISTRY._collector_to_names and not any( + name in REGISTRY._names_to_collectors for name in REGISTRY._get_names(collector) + ): + REGISTRY.register(collector) + + +def _lazy_owner_collectors() -> tuple[Collector, ...]: + SpendLogCleanupMetrics._ensure_initialized() + assert SpendLogCleanupMetrics.rows_deleted is not None + assert SpendLogCleanupMetrics.batch_duration is not None + assert SpendLogCleanupMetrics.rows_remaining is not None + assert SpendLogCleanupMetrics.batch_failures is not None + assert SpendLogCleanupMetrics.runs is not None + in_flight: Final = InFlightRequestsMiddleware._get_gauge() + assert in_flight is not None + breaker: Final = _breaker_metrics() + assert breaker._state_gauge is not None + assert breaker._transitions is not None + assert breaker._failures is not None + return ( + SpendLogCleanupMetrics.rows_deleted, + SpendLogCleanupMetrics.batch_duration, + SpendLogCleanupMetrics.rows_remaining, + SpendLogCleanupMetrics.batch_failures, + SpendLogCleanupMetrics.runs, + in_flight, + breaker._state_gauge, + breaker._transitions, + breaker._failures, + ) + + +def _fresh_admission_collectors() -> tuple[Collector, ...]: + admission: Final = create_prometheus_admission_metrics() + assert admission is not None + return (admission.admitted_gauge, admission.queued_gauge, admission.rejected_counter) + + +@contextmanager +def _isolated_litellm_metric_families(monkeypatch: pytest.MonkeyPatch) -> Iterator[frozenset[str]]: + previous: Final = _registered_collectors() + _unregister_everything() + monkeypatch.setattr(litellm, "prometheus_metrics_config", None) + PrometheusLogger() + PrometheusServicesLogger() + lazy_owned: Final = _lazy_owner_collectors() + _register_if_absent(lazy_owned) + _fresh_admission_collectors() + try: + yield frozenset(metric.name for metric in REGISTRY.collect()) + finally: + _unregister_everything() + for collector in previous: + REGISTRY.register(collector) + _register_if_absent(lazy_owned) + + +@pytest.fixture +def emitted_metric_families(monkeypatch: pytest.MonkeyPatch) -> Iterator[frozenset[str]]: + with _isolated_litellm_metric_families(monkeypatch) as families: + yield families + + +@pytest.fixture +def gauges_registered_by_an_earlier_test() -> Iterator[tuple[Collector, Collector]]: + sentinel: Final = Gauge("litellm_unrelated_sentinel", "registered by a test outside the isolated block") + already_registered: Final = REGISTRY._names_to_collectors.get("litellm_admission_admitted_requests") + admission: Final = already_registered or Gauge( + "litellm_admission_admitted_requests", "registered directly, bypassing admission_control_state" + ) + yield (sentinel, admission) + for gauge in (sentinel,) if already_registered is not None else (sentinel, admission): + if gauge in REGISTRY._collector_to_names: + REGISTRY.unregister(gauge) + + +def test_isolated_metric_families_restore_the_registry_and_keep_lazy_owners_live( + monkeypatch: pytest.MonkeyPatch, gauges_registered_by_an_earlier_test: tuple[Collector, Collector] +): + before: Final = _registered_collectors() + with _isolated_litellm_metric_families(monkeypatch) as families: + assert "litellm_unrelated_sentinel" not in families + assert "litellm_admission_admitted_requests" in families + assert "litellm_in_flight_requests" in families + assert not any(gauge in REGISTRY._collector_to_names for gauge in gauges_registered_by_an_earlier_test) + after: Final = _registered_collectors() + assert all(after[collector] == names for collector, names in before.items()) + lazy_owned: Final = _lazy_owner_collectors() + assert frozenset(after) - frozenset(before) <= frozenset(lazy_owned) + assert all(collector in after for collector in lazy_owned) + + +def _dashboard_expressions(path: Path) -> tuple[str, ...]: + dashboard: Final = json.loads(path.read_text()) + return tuple(target["expr"] for panel in dashboard["panels"] for target in panel.get("targets", ())) + + +def _referenced_metric_tokens(path: Path) -> frozenset[str]: + return frozenset( + token + for expr in _dashboard_expressions(path) + for token in _METRIC_TOKEN_RE.findall(_BY_CLAUSE_RE.sub("", expr)) + ) + + +def _family_of(token: str, families: frozenset[str]) -> str | None: + candidates: Final = (token.removesuffix(suffix) for suffix in _EXPOSITION_SUFFIXES if token.endswith(suffix)) + return next((candidate for candidate in candidates if candidate in families), None) + + +def test_all_metrics_dashboard_charts_every_emitted_metric_family(emitted_metric_families: frozenset[str]): + referenced: Final = _referenced_metric_tokens(_ALL_METRICS_DASHBOARD) + charted: Final = frozenset( + family for token in referenced for family in (_family_of(token, emitted_metric_families),) if family + ) + assert emitted_metric_families - charted == frozenset() + + +@pytest.mark.parametrize("dashboard_path", _LITELLM_DASHBOARDS, ids=lambda p: p.parent.name) +def test_dashboards_only_reference_emitted_metrics(dashboard_path: Path, emitted_metric_families: frozenset[str]): + dead: Final = frozenset( + token + for token in _referenced_metric_tokens(dashboard_path) + if _family_of(token, emitted_metric_families) is None + ) + assert dead == frozenset() + + +@pytest.mark.parametrize("dashboard_path", _LITELLM_DASHBOARDS, ids=lambda p: p.parent.name) +def test_dashboards_use_templated_prometheus_datasource(dashboard_path: Path): + dashboard: Final = json.loads(dashboard_path.read_text()) + datasource_variables: Final = tuple( + variable["name"] for variable in dashboard["templating"]["list"] if variable["type"] == "datasource" + ) + assert datasource_variables == ("DS_PROMETHEUS",) + panel_datasource_uids: Final = frozenset( + panel["datasource"]["uid"] for panel in dashboard["panels"] if panel["type"] != "row" + ) + assert panel_datasource_uids == frozenset({"${DS_PROMETHEUS}"}) def test_remaining_requests_metric_name_in_defined_metrics(): diff --git a/tests/test_litellm/litellm_core_utils/test_litellm_logging.py b/tests/test_litellm/litellm_core_utils/test_litellm_logging.py index aaf44b8e918..e937142e046 100644 --- a/tests/test_litellm/litellm_core_utils/test_litellm_logging.py +++ b/tests/test_litellm/litellm_core_utils/test_litellm_logging.py @@ -7229,3 +7229,155 @@ def test_add_dynamic_callback_registers_once_per_list_without_touching_the_calle assert logging_obj.dynamic_async_failure_callbacks == [callback] assert LitellmLogging._with_dynamic_callback(None, callback) == [callback] assert LitellmLogging._with_dynamic_callback((callback,), callback) == [callback] + + +class TestAzurePTUSpilloverCost: + """Azure PTU deployments price tokens at zero because the reservation is billed flat. + + A request Azure spills onto pay-as-you-go capacity must bill per token instead, so + the zeroed custom pricing has to be skipped when the provider reports spillover. + """ + + ROUTER_MODEL_ID: Final = "ptu-spill-router-model-id" + SERVED_MODEL: Final = "azure/spill-served-model-ptu" + PTU_MODEL_INFO: Final = { + "id": ROUTER_MODEL_ID, + "team_id": "team-1", + "ptu_count": 100, + "cost_per_ptu_per_hour": 1.0, + "ptu_effective_from": "2026-01-01", + "input_cost_per_token": 0.0, + "output_cost_per_token": 0.0, + } + EXPECTED_SPILL_COST: Final = 100 * 2e-6 + 50 * 8e-6 + + @staticmethod + def _register_models() -> None: + litellm.register_model( + model_cost={ + TestAzurePTUSpilloverCost.ROUTER_MODEL_ID: { + "input_cost_per_token": 0.0, + "output_cost_per_token": 0.0, + "litellm_provider": "azure", + "mode": "chat", + }, + TestAzurePTUSpilloverCost.SERVED_MODEL: { + "input_cost_per_token": 2e-6, + "output_cost_per_token": 8e-6, + "litellm_provider": "azure", + "mode": "chat", + }, + } + ) + + @staticmethod + def _unregister_models() -> None: + litellm.model_cost.pop(TestAzurePTUSpilloverCost.ROUTER_MODEL_ID, None) + litellm.model_cost.pop(TestAzurePTUSpilloverCost.SERVED_MODEL, None) + + def _logging_obj(self, model_info: dict, *, flag: str, litellm_rate: float, monkeypatch) -> LitellmLogging: + monkeypatch.setenv("LITELLM_ENABLE_PTU_COST_ATTRIBUTION", flag) + obj = LitellmLogging( + model=self.SERVED_MODEL, + messages=[{"role": "user", "content": "Hi"}], + stream=False, + call_type="completion", + start_time=time.time(), + litellm_call_id="ptu-spill-1", + function_id="f", + ) + obj.update_environment_variables( + model=self.SERVED_MODEL, + user="", + optional_params={}, + litellm_params={ + "api_base": "", + "metadata": {"model_info": model_info}, + "input_cost_per_token": litellm_rate, + "output_cost_per_token": litellm_rate, + }, + custom_llm_provider="azure", + ) + return obj + + @staticmethod + def _response() -> ModelResponse: + from litellm.types.utils import Usage + + return ModelResponse( + id="chatcmpl-spill-1", + created=1234567890, + model="spill-served-model-ptu", + choices=[ + { + "index": 0, + "message": {"role": "assistant", "content": "ok"}, + "finish_reason": "stop", + } + ], + usage=Usage(prompt_tokens=100, completion_tokens=50, total_tokens=150), + ) + + def test_spillover_via_response_additional_headers_bills_per_token(self, monkeypatch) -> None: + self._register_models() + try: + obj = self._logging_obj(dict(self.PTU_MODEL_INFO), flag="True", litellm_rate=0.0, monkeypatch=monkeypatch) + response = self._response() + response._hidden_params["additional_headers"] = {"llm_provider-x-ms-is-spilled-over": "true"} + + assert obj._response_cost_calculator(result=response) == pytest.approx(self.EXPECTED_SPILL_COST) + finally: + self._unregister_models() + + def test_spillover_via_streaming_response_headers_bills_per_token(self, monkeypatch) -> None: + self._register_models() + try: + obj = self._logging_obj(dict(self.PTU_MODEL_INFO), flag="True", litellm_rate=0.0, monkeypatch=monkeypatch) + obj.model_call_details["response_headers"] = { + "x-ms-is-spilled-over": "true", + "x-ms-spillover-from-deployment": "ptu-dep", + } + + assert obj._response_cost_calculator(result=self._response()) == pytest.approx(self.EXPECTED_SPILL_COST) + finally: + self._unregister_models() + + def test_non_spilled_ptu_request_stays_zero_priced(self, monkeypatch) -> None: + self._register_models() + try: + obj = self._logging_obj(dict(self.PTU_MODEL_INFO), flag="True", litellm_rate=0.0, monkeypatch=monkeypatch) + + assert obj._response_cost_calculator(result=self._response()) == 0.0 + finally: + self._unregister_models() + + def test_spillover_header_without_the_flag_stays_zero_priced(self, monkeypatch) -> None: + self._register_models() + try: + obj = self._logging_obj(dict(self.PTU_MODEL_INFO), flag="", litellm_rate=0.0, monkeypatch=monkeypatch) + response = self._response() + response._hidden_params["additional_headers"] = {"llm_provider-x-ms-is-spilled-over": "true"} + + assert obj._response_cost_calculator(result=response) == 0.0 + finally: + self._unregister_models() + + def test_spillover_header_does_not_touch_non_ptu_custom_pricing(self, monkeypatch) -> None: + self._register_models() + custom_model_id: Final = "non-ptu-custom-router-model-id" + litellm.model_cost[custom_model_id] = { + "input_cost_per_token": 1e-6, + "output_cost_per_token": 1e-6, + "litellm_provider": "azure", + "mode": "chat", + } + try: + model_info: Final = {"id": custom_model_id, "input_cost_per_token": 1e-6} + obj = self._logging_obj(model_info, flag="True", litellm_rate=1e-6, monkeypatch=monkeypatch) + response = self._response() + response._hidden_params["additional_headers"] = {"llm_provider-x-ms-is-spilled-over": "true"} + + assert obj._response_cost_calculator(result=response) == pytest.approx(150 * 1e-6) + finally: + litellm.model_cost.pop(custom_model_id, None) + self._unregister_models() diff --git a/tests/test_litellm/litellm_core_utils/test_ptu_pricing.py b/tests/test_litellm/litellm_core_utils/test_ptu_pricing.py index b8fb372d537..1689da2696f 100644 --- a/tests/test_litellm/litellm_core_utils/test_ptu_pricing.py +++ b/tests/test_litellm/litellm_core_utils/test_ptu_pricing.py @@ -7,13 +7,15 @@ from unittest.mock import patch import pytest from litellm.litellm_core_utils.ptu_pricing import ( - ptu_config_error, - ptu_identity_error, CUSTOM_PRICING_FIELDS, PTU_EMPTIED_PRICING_FIELDS, PTU_ZEROED_PRICING_FIELDS, PTU_ZEROED_TABLE_FIELDS, SEARCH_CONTEXT_SIZES, + azure_spillover, + is_spilled_over_ptu_request, + ptu_config_error, + ptu_identity_error, ptu_terms, zeroed_ptu_pricing, ) @@ -294,3 +296,63 @@ def test_an_empty_id_is_no_id(): assert error is not None assert error.startswith("model_info.id is required") + + +def test_the_spillover_header_marks_the_request_as_pay_as_you_go(): + with patch.dict(os.environ, {"LITELLM_ENABLE_PTU_COST_ATTRIBUTION": "True"}, clear=False): + assert ( + is_spilled_over_ptu_request( + model_info=_VALID, + response_headers={"x-ms-is-spilled-over": "True"}, + additional_headers=None, + ) + is True + ) + + +def test_no_spillover_marker_keeps_the_zeroed_ptu_rates(): + with patch.dict(os.environ, {"LITELLM_ENABLE_PTU_COST_ATTRIBUTION": "True"}, clear=False): + assert ( + is_spilled_over_ptu_request( + model_info=_VALID, + response_headers={"x-ms-is-spilled-over": "false"}, + additional_headers=None, + ) + is False + ) + assert ( + is_spilled_over_ptu_request( + model_info=_VALID, + response_headers=None, + additional_headers={"llm_provider-x-ms-is-spilled-over": "absent"}, + ) + is False + ) + + +def test_azure_spillover_carries_the_source_deployment_from_raw_headers(): + assert azure_spillover( + response_headers={ + "x-ms-is-spilled-over": "true", + "x-ms-spillover-from-deployment": "my-ptu", + }, + additional_headers=None, + ) == {"from_deployment": "my-ptu"} + + +def test_azure_spillover_from_processed_headers_has_no_source_when_absent(): + assert azure_spillover( + response_headers=None, + additional_headers={"llm_provider-x-ms-is-spilled-over": "true"}, + ) == {"from_deployment": None} + + +def test_no_spillover_marker_returns_none(): + assert ( + azure_spillover( + response_headers={"x-ms-is-spilled-over": "false"}, + additional_headers=None, + ) + is None + ) + assert azure_spillover(response_headers=None, additional_headers=None) is None diff --git a/tests/test_litellm/llms/azure/test_azure.py b/tests/test_litellm/llms/azure/test_azure.py new file mode 100644 index 00000000000..6b6832f623c --- /dev/null +++ b/tests/test_litellm/llms/azure/test_azure.py @@ -0,0 +1,54 @@ +"""Tests for litellm/llms/azure/azure.py AzureChatCompletion handler behaviour.""" + +import time +from typing import Final + +from openai import AzureOpenAI + +from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj +from litellm.llms.azure.azure import AzureChatCompletion + + +class _FakeRawResponse: + headers: Final = {"x-ms-is-spilled-over": "true"} + + def parse(self): + return iter(()) + + +class _FakeRawCompletions: + def create(self, **kwargs): + return _FakeRawResponse() + + +def test_sync_streaming_stamps_response_headers_on_the_logging_obj() -> None: + """Sync streaming must mirror async_streaming and record the provider response + headers on model_call_details, or downstream consumers (spillover-aware cost + calculation) cannot see them.""" + client = AzureOpenAI(api_key="fake", api_version="2024-02-01", azure_endpoint="https://fake.openai.azure.com") + client.chat.completions.with_raw_response = _FakeRawCompletions() + + logging_obj = LiteLLMLoggingObj( + model="azure/gpt-4o-spill-test", + messages=[{"role": "user", "content": "Hi"}], + stream=True, + call_type="completion", + start_time=time.time(), + litellm_call_id="spill-sync-1", + function_id="f", + ) + + AzureChatCompletion().streaming( + logging_obj=logging_obj, + api_base="https://fake.openai.azure.com", + api_key="fake", + api_version="2024-02-01", + dynamic_params=False, + data={"messages": [{"role": "user", "content": "Hi"}], "stream": True}, + model="gpt-4o-spill-test", + timeout=30.0, + max_retries=0, + client=client, + ) + + assert logging_obj.model_call_details["response_headers"] == {"x-ms-is-spilled-over": "true"} diff --git a/tests/test_litellm/llms/bedrock/chat/test_converse_transformation.py b/tests/test_litellm/llms/bedrock/chat/test_converse_transformation.py index eba8d912fe0..9b28e42f93b 100644 --- a/tests/test_litellm/llms/bedrock/chat/test_converse_transformation.py +++ b/tests/test_litellm/llms/bedrock/chat/test_converse_transformation.py @@ -1189,17 +1189,24 @@ def test_get_supported_openai_params_bedrock_converse(): @pytest.mark.parametrize( - "tools, expected_marker", + "tools, model, expected_marker", [ pytest.param( [{"type": "function", "function": {"name": "f", "parameters": {"type": "object", "properties": {}}}}], + "anthropic.claude-sonnet-4-5-20250929-v1:0", "dep-bedrock", id="tools-present-so-the-cachepoint-is-placed", ), - pytest.param(None, None, id="no-tools-so-nothing-is-placed"), + pytest.param(None, "anthropic.claude-sonnet-4-5-20250929-v1:0", None, id="no-tools-so-nothing-is-placed"), + pytest.param( + [{"type": "function", "function": {"name": "f", "parameters": {"type": "object", "properties": {}}}}], + "global.openai.gpt-6-astra", + None, + id="openai-family-implicit-caching-only", + ), ], ) -def test_tool_config_cachepoint_is_credited_only_where_it_is_placed(tools, expected_marker): +def test_tool_config_cachepoint_is_credited_only_where_it_is_placed(tools, model, expected_marker): """Spend attribution credits the gateway for breakpoints it placed, and a tool_config point becomes one here or nowhere. @@ -1213,7 +1220,7 @@ def test_tool_config_cachepoint_is_credited_only_where_it_is_placed(tools, expec optional_params["tools"] = tools data = AmazonConverseConfig()._transform_request_helper( - model="anthropic.claude-sonnet-4-5-20250929-v1:0", + model=model, system_content_blocks=[], optional_params=optional_params, messages=[{"role": "user", "content": "hi"}], @@ -5591,6 +5598,9 @@ def test_cache_control_injection_tool_config_drops_ttl_for_unsupported_model(): True, id="unmapped-arn-keeps-emitting", ), + pytest.param("global.openai.gpt-6-astra", False, id="openai-family-implicit-caching-only"), + pytest.param("openai.gpt-oss-120b-1:0", False, id="openai-gpt-oss"), + pytest.param("us.openai.gpt-99-unmapped", False, id="unmapped-openai-family-still-suppressed"), ], ) def test_cache_points_emitted_only_for_models_that_support_prompt_caching(model, expects_cache_points, monkeypatch): diff --git a/tests/test_litellm/llms/fireworks_ai/chat/test_fireworks_ai_chat_transformation.py b/tests/test_litellm/llms/fireworks_ai/chat/test_fireworks_ai_chat_transformation.py index f30263bebd5..db25c4307d2 100644 --- a/tests/test_litellm/llms/fireworks_ai/chat/test_fireworks_ai_chat_transformation.py +++ b/tests/test_litellm/llms/fireworks_ai/chat/test_fireworks_ai_chat_transformation.py @@ -978,6 +978,8 @@ def test_llama_vision_supports_vision_from_model_map(): for model in [ "fireworks_ai/accounts/fireworks/models/llama-v3p2-11b-vision-instruct", + "fireworks_ai/accounts/fireworks/models/minimax-m3", + "fireworks_ai/minimax-m3", ]: assert supports_vision(model=model, custom_llm_provider="fireworks_ai") is True assert config.get_provider_info(model)["supports_vision"] is True diff --git a/tests/test_litellm/proxy/auth/test_litellm_license.py b/tests/test_litellm/proxy/auth/test_litellm_license.py index d3f80982c7a..83e26968f97 100644 --- a/tests/test_litellm/proxy/auth/test_litellm_license.py +++ b/tests/test_litellm/proxy/auth/test_litellm_license.py @@ -35,8 +35,8 @@ def test_is_over_limit(): def test_auto_router_capability_limit() -> None: - """Only the signed license's auto_router feature lifts the one-router limit; an API-verified - license (no airgapped data) and an airgapped license without the feature keep it.""" + """The signed license's auto_router feature or its "*" wildcard lifts the one-router limit; an + API-verified license (no airgapped data) and an airgapped license without either keep it.""" license_check = LicenseCheck() license_check.airgapped_license_data = {"expiration_date": "2999-01-01", "allowed_features": ["auto_router"]} assert license_check.auto_router_capability_limit() is None @@ -47,9 +47,18 @@ def test_auto_router_capability_limit() -> None: } assert license_check.auto_router_capability_limit() is None + license_check.airgapped_license_data = {"expiration_date": "2999-01-01", "allowed_features": ["*"]} + assert license_check.auto_router_capability_limit() is None + + license_check.airgapped_license_data = {"expiration_date": "2999-01-01", "allowed_features": ["sso", "*"]} + assert license_check.auto_router_capability_limit() is None + license_check.airgapped_license_data = {"expiration_date": "2999-01-01", "allowed_features": ["sso"]} assert license_check.auto_router_capability_limit() == 1 + license_check.airgapped_license_data = {"expiration_date": "2999-01-01", "allowed_features": "*"} + assert license_check.auto_router_capability_limit() is None + license_check.airgapped_license_data = {"expiration_date": "2999-01-01"} assert license_check.auto_router_capability_limit() == 1 @@ -57,7 +66,9 @@ def test_auto_router_capability_limit() -> None: assert license_check.auto_router_capability_limit() == 1 -def _signed_license(expiration_date: str) -> tuple[RSAPublicKey, str]: +def _signed_license( + expiration_date: str, allowed_features: tuple[str, ...] = ("auto_router",) +) -> tuple[RSAPublicKey, str]: import base64 from cryptography.hazmat.primitives import hashes @@ -65,7 +76,7 @@ def _signed_license(expiration_date: str) -> tuple[RSAPublicKey, str]: private_key = rsa.generate_private_key(public_exponent=65537, key_size=2048) message = json.dumps( - {"expiration_date": expiration_date, "user_id": "u", "allowed_features": ["auto_router"]} + {"expiration_date": expiration_date, "user_id": "u", "allowed_features": list(allowed_features)} ).encode() signature = private_key.sign( message, @@ -99,3 +110,19 @@ def test_valid_signed_license_with_auto_router_lifts_the_limit() -> None: assert license_check.verify_license_without_api_request(public_key=public_key, license_key=license_key) is True assert license_check.auto_router_capability_limit() is None + + +def test_valid_signed_wildcard_license_lifts_the_limit() -> None: + """The license generator defaults allowed_features to ["*"], meaning every feature, so a wildcard + license grants auto_router the same way a license that names it does.""" + license_check = LicenseCheck() + public_key, license_key = _signed_license("2999-01-01", allowed_features=("*",)) + + assert license_check.verify_license_without_api_request(public_key=public_key, license_key=license_key) is True + assert license_check.grants_feature("auto_router") is True + assert license_check.auto_router_capability_limit() is None + + named_public_key, named_key = _signed_license("2999-01-01", allowed_features=("sso", "audit_logs")) + assert license_check.verify_license_without_api_request(public_key=named_public_key, license_key=named_key) is True + assert license_check.grants_feature("auto_router") is False + assert license_check.auto_router_capability_limit() == 1 diff --git a/tests/test_litellm/proxy/client/cli/autoroute/test_commands.py b/tests/test_litellm/proxy/client/cli/autoroute/test_commands.py index 028ab58843f..a46767d8b4f 100644 --- a/tests/test_litellm/proxy/client/cli/autoroute/test_commands.py +++ b/tests/test_litellm/proxy/client/cli/autoroute/test_commands.py @@ -3,13 +3,14 @@ import socket import stat from typing import Optional +import pytest import yaml from click.testing import CliRunner from litellm.proxy.client.cli.commands.claude_settings import ClaudeSettingsError from litellm.proxy.client.cli.commands.autoroute import commands as commands_module from litellm.proxy.client.cli.commands.autoroute import process as process_module -from litellm.proxy.client.cli.commands.autoroute.commands import down, up +from litellm.proxy.client.cli.commands.autoroute.commands import autoroute_group, start, stop from litellm.proxy.client.cli.commands.autoroute.process import PidRecord, ProcessLaunchError, write_pid_record from litellm.proxy.client.cli.commands.up import BackupRecord as ClaudeBackupRecord from litellm.proxy.client.cli.commands.up import write_backup @@ -46,14 +47,14 @@ def _silence_signal_handling(monkeypatch): monkeypatch.setattr(commands_module, "stream_log", lambda *a, **k: None) -class TestUpCommand: +class TestStartCommand: def setup_method(self): self.runner = CliRunner() def test_refuses_when_never_configured(self, monkeypatch, tmp_path): _patch_paths(monkeypatch, tmp_path) - result = self.runner.invoke(up) + result = self.runner.invoke(start) assert result.exit_code != 0 assert "lite autoroute configure" in result.output @@ -66,14 +67,14 @@ class TestUpCommand: config_path.write_text("") monkeypatch.setattr(commands_module, "is_port_available", lambda port: True) - result = self.runner.invoke(up) + result = self.runner.invoke(start) assert result.exit_code != 0 assert result.exception is None or isinstance(result.exception, SystemExit) assert "lite autoroute configure" in result.output def test_refuses_with_actionable_error_when_proxy_runtime_missing(self, monkeypatch, tmp_path): - """`up` launches a real litellm proxy, which the thin `litellm[cli]` install cannot run. + """`start` launches a real litellm proxy, which the thin `litellm[cli]` install cannot run. It must fail fast with an actionable message pointing at the proxy install, before it ever tries to launch the doomed subprocess (which would otherwise die with a bare ImportError).""" config_path, _log_path, _settings_path, _backup_path, _pid_record_path = _patch_paths(monkeypatch, tmp_path) @@ -85,7 +86,7 @@ class TestUpCommand: monkeypatch.setattr(commands_module, "launch_proxy", _fail_if_launched) - result = self.runner.invoke(up) + result = self.runner.invoke(start) assert result.exit_code != 0 assert "fastapi, websockets" in result.output @@ -99,18 +100,18 @@ class TestUpCommand: ) monkeypatch.setattr(commands_module, "is_running", lambda pid: True) - result = self.runner.invoke(up) + result = self.runner.invoke(start) assert result.exit_code != 0 assert "already running" in result.output - assert "lite autoroute down" in result.output + assert "lite autoroute stop" in result.output assert config_path.read_text() == yaml.safe_dump({"model_list": []}) def test_refuses_when_backup_exists_after_an_unclean_crash(self, monkeypatch, tmp_path): - """A prior `up` that was SIGKILL'd leaves no live pid but does leave a stale backup file. + """A prior `start` that was SIGKILL'd leaves no live pid but does leave a stale backup file. - Without this guard, a fresh `up` would overwrite that backup with the currently-patched - (not original) Claude settings, so `down`/Ctrl-C would restore the wrong content forever. + Without this guard, a fresh `start` would overwrite that backup with the currently-patched + (not original) Claude settings, so `stop`/Ctrl-C would restore the wrong content forever. """ config_path, _log_path, claude_settings_path, backup_path, _pid_record_path = _patch_paths( monkeypatch, tmp_path @@ -119,11 +120,11 @@ class TestUpCommand: claude_settings_path.write_text(json.dumps({"env": {"ANTHROPIC_AUTH_TOKEN": "stale-patched-token"}})) write_backup(ClaudeBackupRecord(existed=True, content={"theme": "dark"}), backup_path) - result = self.runner.invoke(up) + result = self.runner.invoke(start) assert result.exit_code != 0 assert "already exists" in result.output - assert "lite autoroute down" in result.output + assert "lite autoroute stop" in result.output assert json.loads(backup_path.read_text())["content"] == {"theme": "dark"} def test_happy_path_patches_settings_then_restores_everything_on_stop(self, monkeypatch, tmp_path): @@ -151,7 +152,7 @@ class TestUpCommand: monkeypatch.setattr("threading.Event.wait", fake_wait) - result = self.runner.invoke(up) + result = self.runner.invoke(start) assert result.exit_code == 0, result.output assert captured["backup_existed"] is True @@ -198,7 +199,7 @@ class TestUpCommand: monkeypatch.setattr("threading.Event.wait", fake_wait) - result = self.runner.invoke(up) + result = self.runner.invoke(start) assert result.exit_code == 0, result.output assert "invalid or unexpected JSON" in result.output @@ -222,7 +223,7 @@ class TestUpCommand: monkeypatch.setattr(commands_module, "terminate", lambda pid, **k: terminate_calls.append(pid)) monkeypatch.setattr(commands_module.secrets, "token_urlsafe", lambda n: "fixed-master-key") - result = self.runner.invoke(up) + result = self.runner.invoke(start) assert result.exit_code != 0 assert "boom" in result.output @@ -234,7 +235,7 @@ class TestUpCommand: def test_terminates_ephemeral_proxy_when_claude_settings_is_corrupt(self, monkeypatch, tmp_path): """The health check can pass and the proxy can come up fine, but if ~/.claude/settings.json turns out to be corrupt, the just-started proxy must not be left - running with no pid record -- exactly the leak `lite autoroute down` exists to clean up.""" + running with no pid record -- exactly the leak `lite autoroute stop` exists to clean up.""" config_path, _log_path, claude_settings_path, backup_path, pid_record_path = _patch_paths(monkeypatch, tmp_path) config_path.write_text(yaml.safe_dump({"model_list": []})) claude_settings_path.write_text("not json at all {{{") @@ -247,7 +248,7 @@ class TestUpCommand: monkeypatch.setattr(commands_module, "terminate", lambda pid, **k: terminate_calls.append(pid)) monkeypatch.setattr(commands_module.secrets, "token_urlsafe", lambda n: "fixed-master-key") - result = self.runner.invoke(up) + result = self.runner.invoke(start) assert result.exit_code != 0 assert "invalid JSON" in result.output @@ -257,7 +258,7 @@ class TestUpCommand: def test_a_status_line_install_failure_leaves_no_backup_behind(self, monkeypatch, tmp_path): # The install runs before the backup is written, so a failure cannot strand a backup that - # would make every later `lite configure` / `lite autoroute up` think a session still owns settings.json + # would make every later `lite configure` / `lite autoroute start` think a session still owns settings.json config_path, _log_path, claude_settings_path, backup_path, pid_record_path = _patch_paths(monkeypatch, tmp_path) config_path.write_text(yaml.safe_dump({"model_list": []})) claude_settings_path.write_text(json.dumps({"theme": "dark"})) @@ -274,7 +275,7 @@ class TestUpCommand: monkeypatch.setattr(commands_module, "install_statusline_script", boom) monkeypatch.setattr(commands_module.secrets, "token_urlsafe", lambda n: "fixed-master-key") - result = self.runner.invoke(up) + result = self.runner.invoke(start) assert result.exit_code != 0 and "disk full" in result.output assert terminate_calls == [778] @@ -282,7 +283,7 @@ class TestUpCommand: assert not backup_path.exists() assert json.loads(claude_settings_path.read_text()) == {"theme": "dark"} - def test_up_uses_the_same_port_and_master_key_across_runs(self, monkeypatch, tmp_path): + def test_start_uses_the_same_port_and_master_key_across_runs(self, monkeypatch, tmp_path): """The LIT-4607/LIT-4608 regression: a client configured against one session must keep working in the next, so consecutive runs must patch settings with an identical base URL and auth token, and the key must be minted exactly once.""" @@ -315,9 +316,9 @@ class TestUpCommand: monkeypatch.setattr("threading.Event.wait", fake_wait) - first = self.runner.invoke(up) + first = self.runner.invoke(start) run_index["current"] = 1 - second = self.runner.invoke(up) + second = self.runner.invoke(start) assert first.exit_code == 0, first.output assert second.exit_code == 0, second.output @@ -326,7 +327,7 @@ class TestUpCommand: assert captured[0]["ANTHROPIC_AUTH_TOKEN"] == captured[1]["ANTHROPIC_AUTH_TOKEN"] assert mint_calls == [32] - def test_up_reuses_a_master_key_already_persisted_in_the_config(self, monkeypatch, tmp_path): + def test_start_reuses_a_master_key_already_persisted_in_the_config(self, monkeypatch, tmp_path): config_path, _log_path, claude_settings_path, _backup_path, _pid_record_path = _patch_paths( monkeypatch, tmp_path ) @@ -354,13 +355,13 @@ class TestUpCommand: monkeypatch.setattr("threading.Event.wait", fake_wait) - result = self.runner.invoke(up) + result = self.runner.invoke(start) assert result.exit_code == 0, result.output assert captured["env"]["ANTHROPIC_AUTH_TOKEN"] == "persisted-key" assert captured["config_text"] == original_config - def test_up_mints_a_fresh_key_when_the_persisted_master_key_is_blank(self, monkeypatch, tmp_path): + def test_start_mints_a_fresh_key_when_the_persisted_master_key_is_blank(self, monkeypatch, tmp_path): config_path, _log_path, claude_settings_path, _backup_path, _pid_record_path = _patch_paths( monkeypatch, tmp_path ) @@ -382,16 +383,22 @@ class TestUpCommand: monkeypatch.setattr("threading.Event.wait", fake_wait) - result = self.runner.invoke(up) + result = self.runner.invoke(start) assert result.exit_code == 0, result.output assert captured["env"]["ANTHROPIC_AUTH_TOKEN"] == "fresh-minted-key" written_config = yaml.safe_load(config_path.read_text()) assert written_config["general_settings"]["master_key"] == "fresh-minted-key" - def test_port_override_reaches_settings_launch_and_pid_record(self, monkeypatch, tmp_path): - """A --port override must flow to every consumer of the port; a hardcoded default in any - one of them would leave the patched settings pointing somewhere the proxy is not.""" + @pytest.mark.parametrize( + ("command", "leading_args"), + [(start, []), (autoroute_group, ["start"]), (autoroute_group, ["up"])], + ids=["start", "group start", "deprecated up alias"], + ) + def test_port_override_reaches_settings_launch_and_pid_record(self, monkeypatch, tmp_path, command, leading_args): + """A --port override must flow to every consumer of the port, through the deprecated `up` + alias too; a hardcoded default in any one of them would leave the patched settings pointing + somewhere the proxy is not.""" config_path, _log_path, claude_settings_path, _backup_path, pid_record_path = _patch_paths( monkeypatch, tmp_path ) @@ -420,16 +427,16 @@ class TestUpCommand: monkeypatch.setattr("threading.Event.wait", fake_wait) - result = self.runner.invoke(up, ["--port", "6111"]) + result = self.runner.invoke(command, [*leading_args, "--port", "6111"]) assert result.exit_code == 0, result.output assert captured["env"]["ANTHROPIC_BASE_URL"] == "http://127.0.0.1:6111" assert launched_ports == [6111] assert captured["pid_record"]["port"] == 6111 - def test_up_rejects_port_4000_which_the_child_proxy_rebinds_unpredictably(self, monkeypatch, tmp_path): + def test_start_rejects_port_4000_which_the_child_proxy_rebinds_unpredictably(self, monkeypatch, tmp_path): """proxy_cli special-cases a busy port 4000 by silently rebinding to a random port, - which would desync base_url from the child; up must refuse 4000 outright.""" + which would desync base_url from the child; start must refuse 4000 outright.""" config_path, _log_path, _settings_path, backup_path, _pid_record_path = _patch_paths(monkeypatch, tmp_path) config_path.write_text(yaml.safe_dump({"model_list": []})) @@ -438,13 +445,13 @@ class TestUpCommand: monkeypatch.setattr(commands_module, "launch_proxy", _fail_launch) - result = self.runner.invoke(up, ["--port", "4000"]) + result = self.runner.invoke(start, ["--port", "4000"]) assert result.exit_code != 0 assert "4000" in result.output assert not backup_path.exists() - def test_up_refuses_when_the_port_is_busy_without_touching_any_state(self, monkeypatch, tmp_path): + def test_start_refuses_when_the_port_is_busy_without_touching_any_state(self, monkeypatch, tmp_path): """A busy port must fail loudly before anything is minted, launched, or patched -- never silently move to another port (the pre-fix behavior this ticket removes).""" config_path, _log_path, claude_settings_path, backup_path, _pid_record_path = _patch_paths( @@ -463,18 +470,18 @@ class TestUpCommand: sock.bind(("127.0.0.1", 0)) sock.listen(1) busy_port = sock.getsockname()[1] - result = self.runner.invoke(up, ["--port", str(busy_port)]) + result = self.runner.invoke(start, ["--port", str(busy_port)]) assert result.exit_code != 0 assert str(busy_port) in result.output - assert "lite autoroute down" in result.output + assert "lite autoroute stop" in result.output assert "--port" in result.output assert config_path.read_text() == original_config assert not backup_path.exists() assert json.loads(claude_settings_path.read_text()) == {"theme": "dark"} -class TestDownCommand: +class TestStopCommand: def setup_method(self): self.runner = CliRunner() @@ -491,7 +498,7 @@ class TestDownCommand: monkeypatch.setattr(commands_module, "is_running", lambda pid: True) monkeypatch.setattr(commands_module, "terminate", lambda pid, **k: terminate_calls.append(pid)) - result = self.runner.invoke(down) + result = self.runner.invoke(stop) assert result.exit_code == 0, result.output assert "Stopped leftover ephemeral proxy" in result.output @@ -501,19 +508,33 @@ class TestDownCommand: assert not backup_path.exists() assert json.loads(claude_settings_path.read_text()) == original_settings + def test_removes_settings_that_did_not_exist_before_start(self, monkeypatch, tmp_path): + _config_path, _log_path, claude_settings_path, backup_path, _pid_record_path = _patch_paths( + monkeypatch, tmp_path + ) + write_backup(ClaudeBackupRecord(existed=False, content=None), backup_path) + claude_settings_path.write_text(json.dumps({"env": {"ANTHROPIC_AUTH_TOKEN": "fixed-master-key"}})) + + result = self.runner.invoke(stop) + + assert result.exit_code == 0, result.output + assert f"Removed {claude_settings_path} (it did not exist before `lite autoroute start`)." in result.output + assert not claude_settings_path.exists() + assert not backup_path.exists() + def test_is_a_clean_no_op_when_nothing_is_running_and_no_backup_exists(self, monkeypatch, tmp_path): _config_path, _log_path, claude_settings_path, _backup_path, _pid_record_path = _patch_paths( monkeypatch, tmp_path ) - result = self.runner.invoke(down) + result = self.runner.invoke(stop) assert result.exit_code == 0, result.output assert "Nothing to restore." in result.output assert not claude_settings_path.exists() def test_clears_a_corrupt_pid_record_and_still_restores_settings(self, monkeypatch, tmp_path): - """down is specifically the crash-recovery path -- a pid file truncated by a mid-write + """stop is specifically the crash-recovery path -- a pid file truncated by a mid-write crash must not block it from clearing the record and restoring Claude settings anyway.""" _config_path, _log_path, claude_settings_path, backup_path, pid_record_path = _patch_paths( monkeypatch, tmp_path @@ -524,7 +545,7 @@ class TestDownCommand: write_backup(ClaudeBackupRecord(existed=True, content=original_settings), backup_path) claude_settings_path.write_text(json.dumps({"env": {"ANTHROPIC_AUTH_TOKEN": "fixed-master-key"}})) - result = self.runner.invoke(down) + result = self.runner.invoke(stop) assert result.exit_code == 0, result.output assert "invalid or unexpected JSON" in result.output @@ -540,7 +561,48 @@ class TestDownCommand: backup_path.parent.mkdir(parents=True, exist_ok=True) backup_path.write_text("not json at all {{{") - result = self.runner.invoke(down) + result = self.runner.invoke(stop) assert result.exit_code != 0 assert "invalid or unexpected JSON" in result.output + + +class TestSubcommandNames: + def test_start_and_stop_are_the_listed_commands(self): + """`lite up` already routes an existing proxy into Claude Code, so the ephemeral proxy's + launcher and its recovery path are listed as `start` and `stop`; the old names stay callable + but are hidden from the listing.""" + runner = CliRunner() + + listing = runner.invoke(autoroute_group, ["--help"]) + assert listing.exit_code == 0, listing.output + listed = {line.split()[0] for line in listing.output.splitlines() if line.startswith(" ")} + assert {"configure", "start", "stop"} <= listed + assert listed.isdisjoint({"up", "down"}) + + for name in ("start", "stop", "up", "down"): + result = runner.invoke(autoroute_group, [name, "--help"]) + assert result.exit_code == 0, result.output + assert "Show this message and exit" in result.output + + def test_up_warns_then_behaves_like_start(self, monkeypatch, tmp_path): + _patch_paths(monkeypatch, tmp_path) + runner = CliRunner() + + result = runner.invoke(autoroute_group, ["up", "--port", "5555"]) + + assert result.exit_code == 1, result.output + assert "`lite autoroute up` is deprecated" in result.stderr + assert "run `lite autoroute start` instead" in result.stderr + assert "No config found. Run `lite autoroute configure` first." in result.output + + def test_down_warns_then_behaves_like_stop(self, monkeypatch, tmp_path): + _patch_paths(monkeypatch, tmp_path) + runner = CliRunner() + + result = runner.invoke(autoroute_group, ["down"]) + + assert result.exit_code == 0, result.output + assert "`lite autoroute down` is deprecated" in result.stderr + assert "run `lite autoroute stop` instead" in result.stderr + assert "Nothing to restore." in result.output diff --git a/tests/test_litellm/proxy/client/cli/test_claude_settings.py b/tests/test_litellm/proxy/client/cli/test_claude_settings.py index cf52d41e963..a48c64eb4a0 100644 --- a/tests/test_litellm/proxy/client/cli/test_claude_settings.py +++ b/tests/test_litellm/proxy/client/cli/test_claude_settings.py @@ -39,7 +39,7 @@ from litellm.proxy.client.cli.commands.claude_settings import ( def _owners(*backup_paths): - """Stand-in owners for the real `lite up` / `lite autoroute up` registry.""" + """Stand-in owners for the real `lite up` / `lite autoroute start` registry.""" return tuple(SettingsFileOwner(path, "lite up", "lite down") for path in backup_paths) @@ -162,7 +162,7 @@ class TestConfigureClaudeSettings: class TestConflictingOwnersOfTheSettingsFile: - """Both `lite up` and `lite autoroute up` restore a backup when they stop. + """Both `lite up` and `lite autoroute start` restore a backup when they stop. Guarding only one of them leaves the other free to silently revert this write, which is the exact hazard the guard exists to prevent. @@ -184,11 +184,11 @@ class TestConflictingOwnersOfTheSettingsFile: settings_path = tmp_path / "claude" / "settings.json" backup = tmp_path / "auto.json" backup.write_text("{}") - autoroute = SettingsFileOwner(backup, "lite autoroute up", "lite autoroute down") + autoroute = SettingsFileOwner(backup, "lite autoroute start", "lite autoroute stop") - with pytest.raises(ClaudeSettingsError, match="`lite autoroute up` is currently managing"): + with pytest.raises(ClaudeSettingsError, match="`lite autoroute start` is currently managing"): _static_configure("https://proxy.example.com", settings_path, (autoroute,)) - with pytest.raises(ClaudeSettingsError, match="Run `lite autoroute down` first"): + with pytest.raises(ClaudeSettingsError, match="Run `lite autoroute stop` first"): _static_configure("https://proxy.example.com", settings_path, (autoroute,)) def test_the_registry_matches_the_paths_the_commands_actually_use(self): @@ -197,7 +197,7 @@ class TestConflictingOwnersOfTheSettingsFile: assert AUTOROUTE_BACKUP_PATH == AUTOROUTE_DIR / "claude_settings_backup.json" assert {o.backup_path for o in SETTINGS_FILE_OWNERS} == {BACKUP_PATH, AUTOROUTE_BACKUP_PATH} - assert {o.stop_command for o in SETTINGS_FILE_OWNERS} == {"lite down", "lite autoroute down"} + assert {o.stop_command for o in SETTINGS_FILE_OWNERS} == {"lite down", "lite autoroute stop"} class TestDoesNotDestroyUserOwnedStructure: @@ -297,7 +297,7 @@ class TestConfigureStatePath: class TestMergeClaudeSettings: - """One merge for every way Claude Code gets wired: `lite up`, `lite configure claude` and `lite autoroute up`.""" + """One merge for every way Claude Code gets wired: `lite up`, `lite configure claude` and `lite autoroute start`.""" def test_a_static_token_lands_in_env_and_the_helper_slot_is_cleared(self): settings = {"apiKeyHelper": "/usr/local/bin/lite auth print-token", "env": {"ANTHROPIC_API_KEY": "leaked"}} @@ -337,7 +337,7 @@ class TestMergeClaudeSettings: def test_a_tier_model_forces_every_claude_code_tier_as_autoroute_needs(self): # Router's auto-router registry is keyed by the literal requested model string with no - # wildcard resolution, so `lite autoroute up` overrides the env var each tier reads. + # wildcard resolution, so `lite autoroute start` overrides the env var each tier reads. settings = {"env": {"ANTHROPIC_DEFAULT_SONNET_MODEL": "claude-opus-4-8"}} merged = merge_claude_settings( settings, "http://127.0.0.1:4000", StaticToken("token-abc"), tier_model="autorouter" diff --git a/tests/test_litellm/proxy/client/cli/test_encryption_commands.py b/tests/test_litellm/proxy/client/cli/test_encryption_commands.py index 43e53cf5be2..3a86eb82593 100644 --- a/tests/test_litellm/proxy/client/cli/test_encryption_commands.py +++ b/tests/test_litellm/proxy/client/cli/test_encryption_commands.py @@ -1,4 +1,4 @@ -"""CLI tests for the ``litellm-proxy encryption migrate`` command. +"""CLI tests for the ``lite encryption migrate`` command. The HTTP client is mocked, so these assert the command's request routing (GET check vs POST migrate, dry-run param) and its residual-state messaging without a diff --git a/tests/test_litellm/proxy/client/cli/test_global_options.py b/tests/test_litellm/proxy/client/cli/test_global_options.py index b73d1acc6e3..d46cc2ad120 100644 --- a/tests/test_litellm/proxy/client/cli/test_global_options.py +++ b/tests/test_litellm/proxy/client/cli/test_global_options.py @@ -1,7 +1,9 @@ # stdlib imports import json import os +import sys from pathlib import Path +from typing import Final from unittest.mock import Mock, patch import pytest @@ -9,7 +11,8 @@ from click.testing import CliRunner import litellm.proxy.client.cli from litellm._version import version as litellm_version -from litellm.proxy.client.cli import cli +from litellm.proxy.client.cli import cli, litellm_proxy_cli +from litellm.proxy.client.cli.main import LITELLM_PROXY_DEPRECATION_NOTICE @pytest.fixture @@ -234,3 +237,32 @@ def test_version_flag_never_sends_api_key_to_unnamed_server(cli_runner, isolated assert all(url.startswith("https://flag-proxy.example.com") for url in requested_urls) sent_keys = [call.kwargs["headers"].get("Authorization") for call in mock_request.call_args_list] assert sent_keys == ["Bearer sk-intended-for-flag-proxy"] * len(requested_urls) + + +def test_litellm_proxy_entrypoint_prints_deprecation_notice_on_stderr_and_still_runs(monkeypatch, capsys, requests_mock): + requests_mock.get("http://localhost:4000/health/readiness", json={"litellm_version": "1.2.3"}) + monkeypatch.setattr(sys, "argv", ["litellm-proxy", "--version"]) + monkeypatch.setenv("LITELLM_PROXY_URL", "http://localhost:4000") + with pytest.raises(SystemExit) as exit_info: + litellm_proxy_cli() + + captured: Final = capsys.readouterr() + assert exit_info.value.code == 0 + assert captured.err.strip() == LITELLM_PROXY_DEPRECATION_NOTICE + assert f"LiteLLM Proxy CLI Version: {litellm_version}" in captured.out + assert "LiteLLM Proxy Server Version: 1.2.3" in captured.out + assert "deprecated" not in captured.out + + +def test_lite_entrypoint_prints_nothing_on_stderr(monkeypatch, capsys, requests_mock): + requests_mock.get("http://localhost:4000/health/readiness", json={"litellm_version": "1.2.3"}) + monkeypatch.setattr(sys, "argv", ["lite", "--version"]) + monkeypatch.setenv("LITELLM_PROXY_URL", "http://localhost:4000") + with pytest.raises(SystemExit) as exit_info: + cli() + + captured: Final = capsys.readouterr() + assert exit_info.value.code == 0 + assert "LiteLLM Proxy Server Version: 1.2.3" in captured.out + assert f"LiteLLM Proxy CLI Version: {litellm_version}" in captured.out + assert captured.err == "" diff --git a/tests/test_litellm/proxy/common_utils/test_upsert_budget_membership.py b/tests/test_litellm/proxy/common_utils/test_upsert_budget_membership.py index e9b4f11e891..2cc0d9f74f5 100644 --- a/tests/test_litellm/proxy/common_utils/test_upsert_budget_membership.py +++ b/tests/test_litellm/proxy/common_utils/test_upsert_budget_membership.py @@ -1,6 +1,6 @@ # tests/litellm/proxy/common_utils/test_upsert_budget_membership.py import types -from datetime import datetime, timezone +from datetime import datetime, timedelta, timezone from unittest.mock import AsyncMock, MagicMock import pytest @@ -27,9 +27,7 @@ def mock_tx(): budget = MagicMock() budget.update = AsyncMock() budget.find_unique = AsyncMock(return_value=None) - budget.create = AsyncMock( - return_value=types.SimpleNamespace(budget_id="new-budget-123") - ) + budget.create = AsyncMock(return_value=types.SimpleNamespace(budget_id="new-budget-123")) tx = MagicMock() tx.litellm_teammembership = membership @@ -83,9 +81,7 @@ async def test_empty_patch_is_noop(mock_tx, fake_user): # member falls back to the team default instead of keeping an empty private row. @pytest.mark.asyncio async def test_clearing_all_limits_disconnects(mock_tx, fake_user): - mock_tx.litellm_budgettable.find_unique = AsyncMock( - return_value=budget_row(max_budget=100.0) - ) + mock_tx.litellm_budgettable.find_unique = AsyncMock(return_value=budget_row(max_budget=100.0)) await _upsert_budget_and_membership( mock_tx, @@ -136,9 +132,7 @@ async def test_clear_one_field_keeps_others(mock_tx, fake_user): # budget_reset_at, so the budget rolls over without waiting for the reset cron. @pytest.mark.asyncio async def test_update_in_place_seeds_reset_at(mock_tx, fake_user): - mock_tx.litellm_budgettable.find_unique = AsyncMock( - return_value=budget_row(max_budget=20.0) - ) + mock_tx.litellm_budgettable.find_unique = AsyncMock(return_value=budget_row(max_budget=20.0)) await _upsert_budget_and_membership( mock_tx, @@ -163,9 +157,7 @@ async def test_update_in_place_seeds_reset_at(mock_tx, fake_user): # budget_duration must not get a (re)computed reset time. @pytest.mark.asyncio async def test_update_in_place_single_field_leaves_reset_at_alone(mock_tx, fake_user): - mock_tx.litellm_budgettable.find_unique = AsyncMock( - return_value=budget_row(max_budget=50.0) - ) + mock_tx.litellm_budgettable.find_unique = AsyncMock(return_value=budget_row(max_budget=50.0)) await _upsert_budget_and_membership( mock_tx, @@ -225,6 +217,7 @@ async def test_create_seeds_reset_at_and_links(mock_tx, fake_user): @pytest.mark.asyncio async def test_clone_on_write_from_shared_default(mock_tx, fake_user): shared_default_id = "team-default-budget-1" + shared_reset_at = datetime.now(timezone.utc) + timedelta(hours=3) mock_tx.litellm_budgettable.find_unique = AsyncMock( return_value=budget_row( budget_id=shared_default_id, @@ -235,6 +228,7 @@ async def test_clone_on_write_from_shared_default(mock_tx, fake_user): rpm_limit=None, model_max_budget=None, budget_duration="1d", + budget_reset_at=shared_reset_at, allowed_models=[], ) ) @@ -252,7 +246,7 @@ async def test_clone_on_write_from_shared_default(mock_tx, fake_user): mock_tx.litellm_budgettable.update.assert_not_called() mock_tx.litellm_budgettable.create.assert_awaited_once() create_data = mock_tx.litellm_budgettable.create.await_args.kwargs["data"] - assert_future_reset_time(create_data.pop("budget_reset_at")) + assert create_data.pop("budget_reset_at") == shared_reset_at assert create_data == { "created_by": fake_user.user_id, "updated_by": fake_user.user_id, @@ -318,9 +312,7 @@ async def test_clone_on_write_clears_duration(mock_tx, fake_user): # team default), we update it in place rather than forking another row. @pytest.mark.asyncio async def test_private_budget_updates_in_place(mock_tx, fake_user): - mock_tx.litellm_budgettable.find_unique = AsyncMock( - return_value=budget_row(max_budget=10.0) - ) + mock_tx.litellm_budgettable.find_unique = AsyncMock(return_value=budget_row(max_budget=10.0)) await _upsert_budget_and_membership( mock_tx, diff --git a/tests/test_litellm/proxy/image_endpoints/test_endpoints.py b/tests/test_litellm/proxy/image_endpoints/test_endpoints.py index 31b87530c94..ad0901e9eee 100644 --- a/tests/test_litellm/proxy/image_endpoints/test_endpoints.py +++ b/tests/test_litellm/proxy/image_endpoints/test_endpoints.py @@ -95,18 +95,14 @@ async def test_image_generation_prompt_rerouting(monkeypatch): monkeypatch.setattr("litellm.proxy.proxy_server.general_settings", {}) monkeypatch.setattr("litellm.proxy.proxy_server.llm_router", None) monkeypatch.setattr("litellm.proxy.proxy_server.proxy_config", {}) - monkeypatch.setattr( - "litellm.proxy.proxy_server.proxy_logging_obj", fake_proxy_logger - ) + monkeypatch.setattr("litellm.proxy.proxy_server.proxy_logging_obj", fake_proxy_logger) monkeypatch.setattr("litellm.proxy.proxy_server.user_model", None) monkeypatch.setattr("litellm.proxy.proxy_server.version", "test-version") monkeypatch.setattr( "litellm.proxy.common_request_processing.ProxyBaseLLMRequestProcessing.get_custom_headers", classmethod(lambda *args, **kwargs: {}), ) - monkeypatch.setattr( - "litellm.proxy.image_endpoints.endpoints.route_request", fake_route_request - ) + monkeypatch.setattr("litellm.proxy.image_endpoints.endpoints.route_request", fake_route_request) result = await endpoints.image_generation( request=request, @@ -141,6 +137,60 @@ def _image_edit_client(monkeypatch, captured: Dict[str, Any]) -> TestClient: return TestClient(app) +def test_image_edit_image_array_alias_is_not_forwarded(monkeypatch): + """The documented `image[]` alias must reach the provider only as `image`.""" + captured: Dict[str, Any] = {} + + response = _image_edit_client(monkeypatch, captured).post( + "/v1/images/edits", + files={"image[]": ("tree.png", b"\x89PNG\r\n\x1a\ntree", "image/png")}, + data={"model": "gpt-image-1", "prompt": "add a hat"}, + ) + + assert response.status_code == 200 + assert "image[]" not in captured + assert [buffer.getvalue() for buffer in captured["image"]] == [b"\x89PNG\r\n\x1a\ntree"] + assert [buffer.name for buffer in captured["image"]] == ["tree.png"] + + +def test_image_edit_mask_array_alias_is_not_forwarded(monkeypatch): + """`mask[]` has the same shape as `image[]` and must be dropped the same way.""" + captured: Dict[str, Any] = {} + + response = _image_edit_client(monkeypatch, captured).post( + "/v1/images/edits", + files={ + "image": ("tree.png", b"\x89PNG\r\n\x1a\ntree", "image/png"), + "mask[]": ("mask.png", b"\x89PNG\r\n\x1a\nmask", "image/png"), + }, + data={"model": "gpt-image-1", "prompt": "add a hat"}, + ) + + assert response.status_code == 200 + assert "mask[]" not in captured + assert [buffer.getvalue() for buffer in captured["mask"]] == [b"\x89PNG\r\n\x1a\nmask"] + assert [buffer.getvalue() for buffer in captured["image"]] == [b"\x89PNG\r\n\x1a\ntree"] + + +def test_image_edit_canonical_file_fields_still_reach_the_provider(monkeypatch): + """Dropping the bracketed aliases must not touch the canonical fields.""" + captured: Dict[str, Any] = {} + + response = _image_edit_client(monkeypatch, captured).post( + "/v1/images/edits", + files={ + "image": ("tree.png", b"\x89PNG\r\n\x1a\ntree", "image/png"), + "mask": ("mask.png", b"\x89PNG\r\n\x1a\nmask", "image/png"), + }, + data={"model": "gpt-image-1", "prompt": "add a hat"}, + ) + + assert response.status_code == 200 + assert [buffer.getvalue() for buffer in captured["image"]] == [b"\x89PNG\r\n\x1a\ntree"] + assert [buffer.getvalue() for buffer in captured["mask"]] == [b"\x89PNG\r\n\x1a\nmask"] + assert captured["prompt"] == "add a hat" + + def test_image_edit_multipart_n_reaches_the_provider_as_an_int(monkeypatch): """A multipart `n` must not arrive as the string Starlette parsed it into.""" captured: Dict[str, Any] = {} @@ -180,7 +230,9 @@ async def test_a_model_the_router_cannot_serve_answers_an_openai_typed_error(mon async def fake_add_litellm_data_to_request(**kwargs: object) -> object: return kwargs["data"] - async def fake_pre_call_hook(*, user_api_key_dict: UserAPIKeyAuth, data: dict[str, object], call_type: str) -> dict[str, object]: + async def fake_pre_call_hook( + *, user_api_key_dict: UserAPIKeyAuth, data: dict[str, object], call_type: str + ) -> dict[str, object]: return data async def fake_post_call_failure_hook(**_: object) -> None: @@ -211,7 +263,9 @@ async def test_a_model_the_router_cannot_serve_answers_an_openai_typed_error(mon request = Request({"type": "http", "method": "POST", "path": "/v1/images/generations", "headers": []}, receive) with pytest.raises(ProxyException) as raised: - await endpoints.image_generation(request=request, fastapi_response=Response(), user_api_key_dict=UserAPIKeyAuth()) + await endpoints.image_generation( + request=request, fastapi_response=Response(), user_api_key_dict=UserAPIKeyAuth() + ) assert (raised.value.type, raised.value.param, raised.value.code) == ("invalid_request_error", None, "404") diff --git a/tests/test_litellm/proxy/management_endpoints/management_v1/test_teams.py b/tests/test_litellm/proxy/management_endpoints/management_v1/test_teams.py new file mode 100644 index 00000000000..9d69f52a834 --- /dev/null +++ b/tests/test_litellm/proxy/management_endpoints/management_v1/test_teams.py @@ -0,0 +1,865 @@ +"""`POST /management/v1/teams/{team_id}/members/bulk_update`: the per-member limit writes and the +HTTP contract around them. + +The in-memory Prisma here follows the one in +`tests/test_litellm/proxy/management_helpers/test_bulk_user_deletion.py`, extended with the budget +table and the membership/budget relation the bulk budget writer needs. +""" + +import copy +import json +from collections.abc import Mapping, Sequence +from contextlib import asynccontextmanager +from datetime import datetime, timedelta, timezone +from typing import Final + +import pytest +from fastapi import FastAPI, Request +from fastapi.exceptions import RequestValidationError +from fastapi.testclient import TestClient +from pydantic import BaseModel, ConfigDict, Field + +from litellm.proxy._types import LiteLLM_TeamTable, LitellmUserRoles, Member, UserAPIKeyAuth +from litellm.proxy.auth.user_api_key_auth import user_api_key_auth +from litellm.proxy.common_utils.user_api_key_cache import ( + UserApiKeyCache, + team_membership_auth_cache_key, + team_membership_reservation_cache_key, +) +from litellm.proxy.db.routing_prisma_wrapper import RoutingPrismaWrapper +from litellm.proxy.list_api.common import ManagementProblem, problem_response, request_validation_problem +from litellm.proxy.management_endpoints.management_v1 import router +from litellm.proxy.management_endpoints.management_v1.common import MANAGEMENT_V1_PREFIX +from litellm.proxy.management_helpers.bulk_team_member_budgets import bulk_update_team_member_budgets +from litellm.types.proxy.management_endpoints.team_endpoints import ( + MAX_BULK_TEAM_MEMBER_BUDGET_UPDATES, + BulkTeamMemberBudgetUpdateRequest, + TeamMemberBudgetUpdateResult, +) + +ADMIN: Final = UserAPIKeyAuth(user_id="admin", user_role=LitellmUserRoles.PROXY_ADMIN, api_key="sk-admin") +OUTSIDER: Final = UserAPIKeyAuth(user_id="outsider", user_role=LitellmUserRoles.INTERNAL_USER) +TEAM_ID: Final = "t1" + + +class _BudgetRow(BaseModel): + """A `LiteLLM_BudgetTable` row, carrying every column the merge patch reads or writes.""" + + model_config = ConfigDict(extra="allow") + + budget_id: str + max_budget: float | None = None + soft_budget: float | None = None + max_parallel_requests: int | None = None + tpm_limit: int | None = None + rpm_limit: int | None = None + model_max_budget: Mapping[str, object] | None = None + budget_duration: str | None = None + budget_reset_at: datetime | None = None + allowed_models: list[str] = Field(default_factory=list) + created_by: str | None = None + updated_by: str | None = None + + +class _MembershipRow(BaseModel): + """A `LiteLLM_TeamMembership` row; `litellm_budget_table` is only filled on an `include` read.""" + + model_config = ConfigDict(extra="allow") + + user_id: str + team_id: str + budget_id: str | None = None + litellm_budget_table: _BudgetRow | None = None + + +def _wanted(where: Mapping[str, object], field: str) -> set[str] | None: + clause: Final = where.get(field) + if isinstance(clause, dict) and "in" in clause: + return set(clause["in"]) + if isinstance(clause, str): + return {clause} + return None + + +def _matches(row: Mapping[str, object], where: Mapping[str, object]) -> bool: + return all((wanted := _wanted(where, field)) is not None and row.get(field) in wanted for field in where) + + +class _BudgetTable: + def __init__(self, budgets: Sequence[_BudgetRow]) -> None: + self.rows: dict[str, _BudgetRow] = {b.budget_id: b for b in budgets} + + async def find_unique(self, where: Mapping[str, str]) -> _BudgetRow | None: + return self.rows.get(where["budget_id"]) + + async def update(self, where: Mapping[str, str], data: Mapping[str, object]) -> _BudgetRow: + row: Final = self.rows[where["budget_id"]] + updated: Final = row.model_copy(update=dict(data)) + self.rows[row.budget_id] = updated + return updated + + async def create(self, data: Mapping[str, object], include: Mapping[str, bool] | None = None) -> _BudgetRow: + budget_id: Final = f"new-budget-{len(self.rows) + 1}" + row: Final = _BudgetRow.model_validate({**data, "budget_id": budget_id}) + self.rows[budget_id] = row + return row + + +class _MembershipTable: + def __init__(self, budgets: _BudgetTable, memberships: Sequence[_MembershipRow]) -> None: + self._budgets = budgets + self.rows: list[_MembershipRow] = list(memberships) + + def _index_of(self, user_id: str, team_id: str) -> int | None: + return next( + (i for i, r in enumerate(self.rows) if r.user_id == user_id and r.team_id == team_id), + None, + ) + + async def find_many( + self, where: Mapping[str, object], include: Mapping[str, bool] | None = None + ) -> list[_MembershipRow]: + matched: Final = [r for r in self.rows if _matches(r.model_dump(), where)] + if not include: + return matched + return [ + r.model_copy(update={"litellm_budget_table": self._budgets.rows.get(r.budget_id or "")}) for r in matched + ] + + async def update(self, where: Mapping[str, Mapping[str, str]], data: Mapping[str, object]) -> _MembershipRow: + key: Final = where["user_id_team_id"] + index: Final = self._index_of(key["user_id"], key["team_id"]) + assert index is not None, f"no membership row for {key}" + relation: Final = data.get("litellm_budget_table") + if isinstance(relation, dict) and relation.get("disconnect"): + self.rows[index] = self.rows[index].model_copy(update={"budget_id": None}) + return self.rows[index] + + async def upsert(self, where: Mapping[str, Mapping[str, str]], data: Mapping[str, object]) -> _MembershipRow: + key: Final = where["user_id_team_id"] + budget_id: Final = data["update"]["litellm_budget_table"]["connect"]["budget_id"] + index: Final = self._index_of(key["user_id"], key["team_id"]) + if index is None: + self.rows.append(_MembershipRow(user_id=key["user_id"], team_id=key["team_id"], budget_id=budget_id)) + return self.rows[-1] + self.rows[index] = self.rows[index].model_copy(update={"budget_id": budget_id}) + return self.rows[index] + + +class _TeamTable: + """`find_many` and `create` are what `RoutingPrismaWrapper` keys read routing off, so a fake + table without them would silently never route and pass a reader-staleness test on the writer.""" + + def __init__(self, teams: Sequence[LiteLLM_TeamTable]) -> None: + self.rows: dict[str, LiteLLM_TeamTable] = {t.team_id: t for t in teams} + + async def find_unique(self, where: Mapping[str, str]) -> LiteLLM_TeamTable | None: + return self.rows.get(where["team_id"]) + + async def find_many(self, where: Mapping[str, object] | None = None) -> list[LiteLLM_TeamTable]: + return [t for t in self.rows.values() if where is None or _matches(t.model_dump(), where)] + + async def create(self, data: Mapping[str, object]) -> LiteLLM_TeamTable: + row: Final = LiteLLM_TeamTable.model_validate(dict(data)) + self.rows[row.team_id] = row + return row + + +class _Db: + def __init__( + self, + teams: Sequence[LiteLLM_TeamTable], + memberships: Sequence[_MembershipRow], + budgets: Sequence[_BudgetRow], + ) -> None: + self.litellm_teamtable = _TeamTable(teams) + self.litellm_budgettable = _BudgetTable(budgets) + self.litellm_teammembership = _MembershipTable(self.litellm_budgettable, memberships) + + +class _FakePrisma: + def __init__( + self, + teams: Sequence[LiteLLM_TeamTable] = (), + memberships: Sequence[_MembershipRow] = (), + budgets: Sequence[_BudgetRow] = (), + ) -> None: + self.db = _Db(teams, memberships, budgets) + + @asynccontextmanager + async def tx(self, *, timeout: object = None): + snapshot: Final = copy.deepcopy(self.db) + try: + yield self.db + except BaseException: + self.db = snapshot + raise + + +class _ReplicatedPrisma: + """A client whose reads route to a lagging replica, as a proxy with `DATABASE_URL_READ_REPLICA` does.""" + + def __init__(self, writer: _FakePrisma, reader: _FakePrisma) -> None: + self._writer = writer + self.db = RoutingPrismaWrapper(writer=writer.db, reader=reader.db) # pyright: ignore[reportArgumentType] # fake dbs stand in for PrismaWrapper + + def tx(self, *, timeout: object = None): + return self._writer.tx(timeout=timeout) + + +class _UnreachableDb: + """A `.db` whose every table access fails, as one behind a dropped connection does.""" + + def __getattr__(self, name: str) -> object: + raise RuntimeError("connection reset by peer") + + +class _UnreachablePrisma: + def __init__(self) -> None: + self.db = _UnreachableDb() + + +def _team( + *members: str, + team_id: str = TEAM_ID, + default_budget_id: str | None = None, + admins: Sequence[str] = (), +) -> LiteLLM_TeamTable: + return LiteLLM_TeamTable( + team_id=team_id, + metadata={"team_member_budget_id": default_budget_id} if default_budget_id else {}, + members_with_roles=[ + Member(user_id=m, user_email=f"{m}@example.com", role="admin" if m in admins else "user") for m in members + ], + ) + + +def _membership(user_id: str, budget_id: str | None = None, team_id: str = TEAM_ID) -> _MembershipRow: + return _MembershipRow(user_id=user_id, team_id=team_id, budget_id=budget_id) + + +def _budget( + budget_id: str, + *, + max_budget: float | None = None, + tpm_limit: int | None = None, + rpm_limit: int | None = None, + budget_duration: str | None = None, + budget_reset_at: datetime | None = None, +) -> _BudgetRow: + return _BudgetRow( + budget_id=budget_id, + max_budget=max_budget, + tpm_limit=tpm_limit, + rpm_limit=rpm_limit, + budget_duration=budget_duration, + budget_reset_at=budget_reset_at, + ) + + +async def _bulk_update( + prisma: _FakePrisma | _ReplicatedPrisma, + members: Sequence[Mapping[str, object]], + team_id: str = TEAM_ID, + caller: UserAPIKeyAuth = ADMIN, + cache: UserApiKeyCache | None = None, +) -> tuple[TeamMemberBudgetUpdateResult, ...]: + return await bulk_update_team_member_budgets( + team_id=team_id, + data=BulkTeamMemberBudgetUpdateRequest.model_validate({"members": list(members)}), + user_api_key_dict=caller, + prisma_client=prisma, # pyright: ignore[reportArgumentType] # fake stands in for PrismaClient + user_api_key_cache=cache or UserApiKeyCache(), + litellm_proxy_admin_name="default_user_id", + ) + + +def _budget_id_of(prisma: _FakePrisma, user_id: str, team_id: str = TEAM_ID) -> str | None: + row: Final = next(r for r in prisma.db.litellm_teammembership.rows if r.user_id == user_id and r.team_id == team_id) + return row.budget_id + + +def _budget_of(prisma: _FakePrisma, user_id: str, team_id: str = TEAM_ID) -> _BudgetRow: + budget_id: Final = _budget_id_of(prisma, user_id, team_id) + assert budget_id is not None, f"{user_id} has no budget" + return prisma.db.litellm_budgettable.rows[budget_id] + + +def _seeded_cache(*user_ids: str, team_id: str = TEAM_ID) -> UserApiKeyCache: + cache: Final = UserApiKeyCache() + for user_id in user_ids: + cache.set_cache(key=team_membership_auth_cache_key(team_id=team_id, user_id=user_id), value={"cap": "old"}) + cache.set_cache( + key=team_membership_reservation_cache_key(user_id=user_id, team_id=team_id), value={"cap": "old"} + ) + return cache + + +def _cached_keys(cache: UserApiKeyCache, user_id: str, team_id: str = TEAM_ID) -> tuple[object, object]: + return ( + cache.get_cache(key=team_membership_auth_cache_key(team_id=team_id, user_id=user_id)), + cache.get_cache(key=team_membership_reservation_cache_key(user_id=user_id, team_id=team_id)), + ) + + +@pytest.mark.asyncio +async def test_patching_one_member_of_a_shared_budget_row_forks_it_and_leaves_the_other_member_untouched(): + prisma = _FakePrisma( + teams=[_team("m1", "m2")], + memberships=[_membership("m1", "shared-b"), _membership("m2", "shared-b")], + budgets=[_budget("shared-b", max_budget=100.0, tpm_limit=900)], + ) + + results = await _bulk_update(prisma, [{"user_id": "m1", "max_budget_in_team": 50}]) + + assert [(r.user_id, r.success, r.max_budget) for r in results] == [("m1", True, 50.0)] + assert _budget_id_of(prisma, "m1") not in (None, "shared-b") + assert (_budget_of(prisma, "m1").max_budget, _budget_of(prisma, "m1").tpm_limit) == (50.0, 900) + assert _budget_id_of(prisma, "m2") == "shared-b" + assert prisma.db.litellm_budgettable.rows["shared-b"].max_budget == 100.0 + assert results[0].budget_id == _budget_id_of(prisma, "m1") + + +@pytest.mark.asyncio +async def test_patching_members_of_the_team_default_budget_gives_each_their_own_row_and_leaves_the_default_alone(): + prisma = _FakePrisma( + teams=[_team("m1", "m2", "m3", default_budget_id="team-default")], + memberships=[ + _membership("m1", "team-default"), + _membership("m2", "team-default"), + _membership("m3", "team-default"), + ], + budgets=[_budget("team-default", max_budget=25.0, tpm_limit=1000)], + ) + + results = await _bulk_update( + prisma, + [{"user_id": "m1", "max_budget_in_team": 5}, {"user_id": "m2", "max_budget_in_team": 7}], + ) + + assert [r.success for r in results] == [True, True] + default = prisma.db.litellm_budgettable.rows["team-default"] + assert (default.max_budget, default.tpm_limit) == (25.0, 1000) + assert _budget_id_of(prisma, "m3") == "team-default" + patched = (_budget_id_of(prisma, "m1"), _budget_id_of(prisma, "m2")) + assert len(set(patched)) == 2 and "team-default" not in patched + assert (_budget_of(prisma, "m1").max_budget, _budget_of(prisma, "m1").tpm_limit) == (5.0, 1000) + assert (_budget_of(prisma, "m2").max_budget, _budget_of(prisma, "m2").tpm_limit) == (7.0, 1000) + + +@pytest.mark.asyncio +async def test_the_team_default_row_is_forked_even_when_only_one_membership_points_at_it(): + prisma = _FakePrisma( + teams=[_team("m1", "m2", default_budget_id="team-default")], + memberships=[_membership("m1", "team-default")], + budgets=[_budget("team-default", max_budget=25.0, tpm_limit=1000)], + ) + + results = await _bulk_update(prisma, [{"user_id": "m1", "max_budget_in_team": 5}]) + + assert [(r.success, r.max_budget, r.tpm_limit) for r in results] == [(True, 5.0, 1000)] + default = prisma.db.litellm_budgettable.rows["team-default"] + assert (default.max_budget, default.tpm_limit) == (25.0, 1000) + assert _budget_id_of(prisma, "m1") not in (None, "team-default") + + +@pytest.mark.asyncio +async def test_a_budget_row_only_one_member_points_at_is_updated_in_place(): + prisma = _FakePrisma( + teams=[_team("m1", "m2", default_budget_id="team-default")], + memberships=[_membership("m1", "priv-m1"), _membership("m2", "team-default")], + budgets=[_budget("team-default", max_budget=25.0), _budget("priv-m1", max_budget=10.0, tpm_limit=5)], + ) + + results = await _bulk_update(prisma, [{"user_id": "m1", "max_budget_in_team": 20}]) + + assert [(r.success, r.budget_id, r.max_budget) for r in results] == [(True, "priv-m1", 20.0)] + assert set(prisma.db.litellm_budgettable.rows) == {"team-default", "priv-m1"} + assert _budget_id_of(prisma, "m1") == "priv-m1" + assert (_budget_of(prisma, "m1").max_budget, _budget_of(prisma, "m1").tpm_limit) == (20.0, 5) + + +@pytest.mark.asyncio +async def test_an_omitted_field_is_kept_an_explicit_null_clears_it_and_clearing_the_last_limit_disconnects(): + prisma = _FakePrisma( + teams=[_team("m1")], + memberships=[_membership("m1", "priv-m1")], + budgets=[_budget("priv-m1", max_budget=10.0, tpm_limit=5, rpm_limit=7)], + ) + + kept = await _bulk_update(prisma, [{"user_id": "m1", "rpm_limit": 9}]) + + assert (kept[0].max_budget, kept[0].tpm_limit, kept[0].rpm_limit) == (10.0, 5, 9) + + cleared = await _bulk_update(prisma, [{"user_id": "m1", "tpm_limit": None}]) + + assert (cleared[0].max_budget, cleared[0].tpm_limit, cleared[0].rpm_limit) == (10.0, None, 9) + assert _budget_id_of(prisma, "m1") == "priv-m1" + + emptied = await _bulk_update(prisma, [{"user_id": "m1", "max_budget_in_team": None, "rpm_limit": None}]) + + assert (emptied[0].success, emptied[0].budget_id, emptied[0].max_budget) == (True, None, None) + assert _budget_id_of(prisma, "m1") is None + + +@pytest.mark.asyncio +async def test_budget_duration_seeds_a_reset_time_derived_from_the_duration_and_clearing_it_clears_the_reset(): + prisma = _FakePrisma( + teams=[_team("m1", "m2")], + memberships=[_membership("m1", "priv-m1"), _membership("m2", "priv-m2")], + budgets=[_budget("priv-m1", max_budget=10.0), _budget("priv-m2", max_budget=10.0)], + ) + before = datetime.now(timezone.utc) + + await _bulk_update( + prisma, + [{"user_id": "m1", "budget_duration": "2d"}, {"user_id": "m2", "budget_duration": "5d"}], + ) + + two_day = _budget_of(prisma, "m1").budget_reset_at + five_day = _budget_of(prisma, "m2").budget_reset_at + assert two_day is not None and five_day is not None + assert before < two_day <= before + timedelta(days=2) + assert before + timedelta(days=4) - timedelta(seconds=1) < five_day <= before + timedelta(days=5) + assert five_day - two_day == timedelta(days=3) + + await _bulk_update(prisma, [{"user_id": "m1", "budget_duration": None}]) + + assert _budget_of(prisma, "m1").budget_reset_at is None + assert _budget_of(prisma, "m1").budget_duration is None + assert _budget_of(prisma, "m1").max_budget == 10.0 + + +@pytest.mark.asyncio +async def test_a_member_named_twice_is_written_once_and_the_later_rows_report_the_duplicate(): + prisma = _FakePrisma( + teams=[_team("m1")], + memberships=[_membership("m1", "priv-m1")], + budgets=[_budget("priv-m1", max_budget=1.0)], + ) + + results = await _bulk_update( + prisma, + [ + {"user_id": "m1", "max_budget_in_team": 10}, + {"user_id": "m1", "max_budget_in_team": 20}, + {"user_email": "m1@example.com", "max_budget_in_team": 30}, + ], + ) + + assert [(r.success, r.error) for r in results] == [ + (True, None), + (False, "Duplicate member in request"), + (False, "Duplicate member in request"), + ] + assert _budget_of(prisma, "m1").max_budget == 10.0 + + +@pytest.mark.asyncio +async def test_a_row_naming_somebody_off_the_team_fails_without_writing_while_the_rest_of_the_batch_lands(): + prisma = _FakePrisma( + teams=[_team("m1")], + memberships=[_membership("m1", "priv-m1"), _membership("elsewhere", "priv-other")], + budgets=[_budget("priv-m1", max_budget=1.0), _budget("priv-other", max_budget=2.0)], + ) + + results = await _bulk_update( + prisma, + [ + {"user_id": "elsewhere", "max_budget_in_team": 99}, + {"user_email": "nobody@example.com", "max_budget_in_team": 99}, + {"user_id": "m1", "max_budget_in_team": 10}, + ], + ) + + assert [(r.success, r.error) for r in results] == [ + (False, "User not found in team"), + (False, "User not found in team"), + (True, None), + ] + assert prisma.db.litellm_budgettable.rows["priv-other"].max_budget == 2.0 + assert _budget_of(prisma, "m1").max_budget == 10.0 + assert set(prisma.db.litellm_budgettable.rows) == {"priv-m1", "priv-other"} + + +@pytest.mark.asyncio +async def test_each_result_carries_the_limits_read_back_after_the_write_in_request_order(): + prisma = _FakePrisma( + teams=[_team("m1", "m2")], + memberships=[_membership("m1", "priv-m1"), _membership("m2", "priv-m2")], + budgets=[ + _budget("priv-m1", tpm_limit=100, budget_duration="7d"), + _budget("priv-m2", rpm_limit=3), + ], + ) + + results = await _bulk_update( + prisma, + [{"user_id": "m2", "rpm_limit": 8}, {"user_id": "m1", "max_budget_in_team": 42}], + ) + + assert [r.user_id for r in results] == ["m2", "m1"] + assert (results[1].max_budget, results[1].tpm_limit, results[1].budget_duration) == (42.0, 100, "7d") + assert (results[0].rpm_limit, results[0].max_budget) == (8, None) + + +@pytest.mark.asyncio +async def test_every_written_member_is_evicted_from_both_team_membership_cache_keys(): + prisma = _FakePrisma( + teams=[_team("m1", "m2", "m3")], + memberships=[_membership("m1", "priv-m1"), _membership("m2", "priv-m2"), _membership("m3", "priv-m3")], + budgets=[_budget("priv-m1", max_budget=1.0), _budget("priv-m2", max_budget=2.0), _budget("priv-m3")], + ) + cache = _seeded_cache("m1", "m2", "m3") + + await _bulk_update( + prisma, + [{"user_id": "m1", "max_budget_in_team": 10}, {"user_id": "m2", "max_budget_in_team": 20}], + cache=cache, + ) + + assert _cached_keys(cache, "m1") == (None, None) + assert _cached_keys(cache, "m2") == (None, None) + assert _cached_keys(cache, "m3") == ({"cap": "old"}, {"cap": "old"}) + + +@pytest.mark.asyncio +async def test_a_member_with_no_cap_of_their_own_reports_the_team_default_cap_but_only_their_own_rate_limits(): + prisma = _FakePrisma( + teams=[_team("m1", default_budget_id="team-default")], + memberships=[], + budgets=[_budget("team-default", max_budget=25.0, tpm_limit=1000)], + ) + + results = await _bulk_update(prisma, [{"user_id": "m1", "tpm_limit": 7}]) + + assert [(r.success, r.max_budget, r.max_budget_source, r.tpm_limit) for r in results] == [ + (True, 25.0, "team_default", 7) + ] + assert _budget_of(prisma, "m1").max_budget is None + default = prisma.db.litellm_budgettable.rows["team-default"] + assert (default.max_budget, default.tpm_limit) == (25.0, 1000) + + +@pytest.mark.asyncio +async def test_an_explicit_cap_reports_as_the_members_own_while_clearing_one_falls_back_to_the_team_default(): + prisma = _FakePrisma( + teams=[_team("m1", "m2", default_budget_id="team-default")], + memberships=[_membership("m1", "priv-m1"), _membership("m2", "priv-m2")], + budgets=[ + _budget("team-default", max_budget=25.0), + _budget("priv-m1", max_budget=5.0), + _budget("priv-m2", max_budget=9.0), + ], + ) + + results = await _bulk_update( + prisma, + [{"user_id": "m1", "max_budget_in_team": 50}, {"user_id": "m2", "max_budget_in_team": None}], + ) + + assert [(r.user_id, r.max_budget, r.max_budget_source) for r in results] == [ + ("m1", 50.0, "member"), + ("m2", 25.0, "team_default"), + ] + assert results[1].budget_id is None + assert _budget_id_of(prisma, "m2") is None + assert prisma.db.litellm_budgettable.rows["team-default"].max_budget == 25.0 + + +@pytest.mark.asyncio +async def test_a_team_with_no_default_budget_reports_no_effective_cap_for_a_member_without_one(): + prisma = _FakePrisma( + teams=[_team("m1")], + memberships=[_membership("m1", "priv-m1")], + budgets=[_budget("priv-m1", tpm_limit=5)], + ) + + results = await _bulk_update(prisma, [{"user_id": "m1", "rpm_limit": 3}]) + + assert [(r.success, r.max_budget, r.max_budget_source) for r in results] == [(True, None, None)] + assert (results[0].tpm_limit, results[0].rpm_limit) == (5, 3) + + +@pytest.mark.asyncio +async def test_a_zero_team_default_reports_no_cap_because_enforcement_reads_zero_there_as_uncapped(): + prisma = _FakePrisma( + teams=[_team("m1", default_budget_id="team-default")], + memberships=[_membership("m1", None)], + budgets=[_budget("team-default", max_budget=0.0)], + ) + + results = await _bulk_update(prisma, [{"user_id": "m1", "tpm_limit": 9}]) + + assert [(r.success, r.max_budget, r.max_budget_source) for r in results] == [(True, None, None)] + assert results[0].tpm_limit == 9 + + +@pytest.mark.asyncio +async def test_a_row_that_names_nobody_on_the_team_reports_no_cap_and_no_source(): + prisma = _FakePrisma( + teams=[_team("m1", default_budget_id="team-default")], + memberships=[_membership("m1", "priv-m1")], + budgets=[_budget("team-default", max_budget=25.0), _budget("priv-m1", max_budget=5.0)], + ) + + results = await _bulk_update( + prisma, + [{"user_id": "ghost", "max_budget_in_team": 1}, {"user_id": "m1", "max_budget_in_team": 6}], + ) + + assert [(r.success, r.max_budget, r.max_budget_source) for r in results] == [ + (False, None, None), + (True, 6.0, "member"), + ] + + +@pytest.mark.asyncio +async def test_the_roster_authz_read_runs_on_the_writer_so_a_lagging_replica_cannot_let_a_demoted_admin_write(): + writer = _FakePrisma( + teams=[_team("lead", "m1")], + memberships=[_membership("m1", "priv-m1")], + budgets=[_budget("priv-m1", max_budget=1.0)], + ) + replica = _FakePrisma(teams=[_team("lead", "m1", admins=("lead",))]) + demoted = UserAPIKeyAuth(user_id="lead", user_role=LitellmUserRoles.INTERNAL_USER) + + with pytest.raises(ManagementProblem) as raised: + await _bulk_update( + _ReplicatedPrisma(writer=writer, reader=replica), + [{"user_id": "m1", "max_budget_in_team": 99}], + caller=demoted, + ) + + assert raised.value.problem.status == 403 + assert writer.db.litellm_budgettable.rows["priv-m1"].max_budget == 1.0 + + +@pytest.mark.asyncio +async def test_the_batch_writes_one_audit_entry_carrying_every_written_members_limits_before_and_after(monkeypatch): + import litellm + from litellm.proxy._types import LitellmTableNames + + monkeypatch.setattr(litellm, "store_audit_logs", True) + captured: list[object] = [] + + async def capture(request_data): + captured.append(request_data) + + monkeypatch.setattr("litellm.proxy.management_helpers.audit_logs.create_audit_log_for_update", capture) + prisma = _FakePrisma( + teams=[_team("m1", "m2")], + memberships=[_membership("m1", "priv-m1"), _membership("m2", "priv-m2")], + budgets=[_budget("priv-m1", max_budget=1.0), _budget("priv-m2", max_budget=2.0)], + ) + + await _bulk_update(prisma, [{"user_id": "m1", "max_budget_in_team": 10}]) + + assert len(captured) == 1 + entry = captured[0] + assert (entry.object_id, entry.action, entry.table_name) == ( + TEAM_ID, + "updated", + LitellmTableNames.TEAM_TABLE_NAME, + ) + before = {row["user_id"]: row for row in json.loads(entry.before_value)["team_member_budgets"]} + after = {row["user_id"]: row for row in json.loads(entry.updated_values)["team_member_budgets"]} + assert (before["m1"]["max_budget"], after["m1"]["max_budget"]) == (1.0, 10.0) + assert "m2" not in before and "m2" not in after + + +@pytest.mark.asyncio +async def test_no_audit_entry_is_written_when_audit_logging_is_off(monkeypatch): + import litellm + + monkeypatch.setattr(litellm, "store_audit_logs", False) + captured: list[object] = [] + + async def capture(request_data): + captured.append(request_data) + + monkeypatch.setattr("litellm.proxy.management_helpers.audit_logs.create_audit_log_for_update", capture) + prisma = _FakePrisma( + teams=[_team("m1")], + memberships=[_membership("m1", "priv-m1")], + budgets=[_budget("priv-m1", max_budget=1.0)], + ) + + await _bulk_update(prisma, [{"user_id": "m1", "max_budget_in_team": 10}]) + + assert captured == [] + assert _budget_of(prisma, "m1").max_budget == 10.0 + + +@pytest.mark.asyncio +async def test_forking_a_shared_row_keeps_its_reset_window_so_an_unrelated_limit_edit_grants_no_free_period(): + shared_reset_at = datetime.now(timezone.utc) + timedelta(days=3) + prisma = _FakePrisma( + teams=[_team("m1", "m2")], + memberships=[_membership("m1", "shared-b"), _membership("m2", "shared-b")], + budgets=[_budget("shared-b", max_budget=100.0, budget_duration="30d", budget_reset_at=shared_reset_at)], + ) + + results = await _bulk_update(prisma, [{"user_id": "m1", "tpm_limit": 9}]) + + assert [(r.success, r.budget_duration) for r in results] == [(True, "30d")] + assert _budget_id_of(prisma, "m1") not in (None, "shared-b") + assert _budget_of(prisma, "m1").budget_reset_at == shared_reset_at + assert prisma.db.litellm_budgettable.rows["shared-b"].budget_reset_at == shared_reset_at + + +@pytest.mark.asyncio +async def test_forking_a_shared_row_does_restart_the_window_when_the_patch_sets_a_new_duration(): + shared_reset_at = datetime.now(timezone.utc) + timedelta(days=3) + prisma = _FakePrisma( + teams=[_team("m1", "m2")], + memberships=[_membership("m1", "shared-b"), _membership("m2", "shared-b")], + budgets=[_budget("shared-b", max_budget=100.0, budget_duration="30d", budget_reset_at=shared_reset_at)], + ) + + await _bulk_update(prisma, [{"user_id": "m1", "budget_duration": "1d"}]) + + forked = _budget_of(prisma, "m1").budget_reset_at + assert forked is not None and forked != shared_reset_at + assert forked <= datetime.now(timezone.utc) + timedelta(days=1) + + +app = FastAPI() + + +@app.exception_handler(ManagementProblem) +async def management_problem_exception_handler(request: Request, exc: ManagementProblem): + return problem_response(exc.problem) + + +@app.exception_handler(RequestValidationError) +async def validation_exception_handler(request: Request, exc: RequestValidationError): + return problem_response(request_validation_problem(exc.errors())) + + +app.include_router(router) +client = TestClient(app) + +BULK_UPDATE_PATH: Final = f"{MANAGEMENT_V1_PREFIX}/teams/{TEAM_ID}/members/bulk_update" + + +@pytest.fixture +def as_proxy_admin(): + app.dependency_overrides[user_api_key_auth] = lambda: ADMIN + yield + app.dependency_overrides.clear() + + +@pytest.fixture +def as_outsider(): + app.dependency_overrides[user_api_key_auth] = lambda: OUTSIDER + yield + app.dependency_overrides.clear() + + +@pytest.fixture +def prisma(monkeypatch): + fake = _FakePrisma( + teams=[_team("m1", "m2")], + memberships=[_membership("m1", "priv-m1")], + budgets=[_budget("priv-m1", max_budget=1.0)], + ) + monkeypatch.setattr("litellm.proxy.proxy_server.prisma_client", fake) + return fake + + +def _post(body: object, path: str = BULK_UPDATE_PATH): + return client.post(path, json=body, headers={"Authorization": "Bearer sk-1234"}) + + +def test_unknown_fields_empty_and_oversized_batches_are_422_problem_documents(prisma, as_proxy_admin): + bodies = ( + {"members": [{"user_id": "m1", "max_budget": 10}]}, + {"members": [{"user_id": "m1"}], "team_id": TEAM_ID}, + {"members": []}, + {"members": [{"user_id": f"u{i}"} for i in range(MAX_BULK_TEAM_MEMBER_BUDGET_UPDATES + 1)]}, + ) + + for body in bodies: + response = _post(body) + + assert response.status_code == 422, body + assert response.headers["content-type"] == "application/problem+json" + assert response.json()["type"] == "urn:litellm:error:invalid-request-body" + assert prisma.db.litellm_budgettable.rows["priv-m1"].max_budget == 1.0 + + +def test_an_unknown_team_is_a_404_problem_document(prisma, as_proxy_admin): + response = _post( + {"members": [{"user_id": "m1", "max_budget_in_team": 10}]}, + path=f"{MANAGEMENT_V1_PREFIX}/teams/nope/members/bulk_update", + ) + + assert response.status_code == 404 + assert response.headers["content-type"] == "application/problem+json" + assert response.json()["type"] == "urn:litellm:error:team-not-found" + assert prisma.db.litellm_budgettable.rows["priv-m1"].max_budget == 1.0 + + +def test_a_caller_who_administers_neither_the_team_nor_its_org_is_a_403_problem_document(prisma, as_outsider): + response = _post({"members": [{"user_id": "m1", "max_budget_in_team": 10}]}) + + assert response.status_code == 403 + assert response.headers["content-type"] == "application/problem+json" + assert response.json()["type"] == "urn:litellm:error:forbidden" + assert prisma.db.litellm_budgettable.rows["priv-m1"].max_budget == 1.0 + + +def test_a_team_admin_may_bulk_update_their_own_teams_members(prisma, monkeypatch): + prisma.db.litellm_teamtable.rows[TEAM_ID] = _team("lead", "m1", admins=("lead",)) + app.dependency_overrides[user_api_key_auth] = lambda: UserAPIKeyAuth( + user_id="lead", user_role=LitellmUserRoles.INTERNAL_USER + ) + try: + response = _post({"members": [{"user_id": "m1", "max_budget_in_team": 10}]}) + finally: + app.dependency_overrides.clear() + + assert response.status_code == 200 + assert [(r["user_id"], r["success"], r["max_budget"]) for r in response.json()["data"]] == [("m1", True, 10.0)] + + +@pytest.mark.parametrize("duration", ("0d", "nonsense")) +def test_a_budget_duration_no_reset_can_be_scheduled_from_is_a_422_naming_its_row_and_writes_nothing( + prisma, as_proxy_admin, duration +): + response = _post( + { + "members": [ + {"user_id": "m1", "max_budget_in_team": 10}, + {"user_id": "m2", "budget_duration": duration}, + ] + } + ) + + assert response.status_code == 422 + assert response.headers["content-type"] == "application/problem+json" + assert response.json()["type"] == "urn:litellm:error:invalid-request-body" + assert "members.1.budget_duration" in response.json()["detail"] + assert prisma.db.litellm_budgettable.rows["priv-m1"].max_budget == 1.0 + + +def test_an_unconnected_database_is_a_503_problem_document(monkeypatch, as_proxy_admin): + monkeypatch.setattr("litellm.proxy.proxy_server.prisma_client", None) + + response = _post({"members": [{"user_id": "m1", "max_budget_in_team": 10}]}) + + assert response.status_code == 503 + assert response.headers["content-type"] == "application/problem+json" + assert response.json()["type"] == "urn:litellm:error:database-not-connected" + + +def test_a_driver_error_answers_as_a_problem_document_without_leaking_the_exception(monkeypatch, as_proxy_admin): + monkeypatch.setattr("litellm.proxy.proxy_server.prisma_client", _UnreachablePrisma()) + + response = _post({"members": [{"user_id": "m1", "max_budget_in_team": 10}]}) + + assert response.status_code == 500 + assert response.headers["content-type"] == "application/problem+json" + assert response.json()["type"] == "urn:litellm:error:internal-server-error" + assert "connection reset by peer" not in response.text diff --git a/tests/test_litellm/proxy/management_endpoints/scim/test_scim_v2_endpoints.py b/tests/test_litellm/proxy/management_endpoints/scim/test_scim_v2_endpoints.py index 60f9a1a55e2..364ec4aad61 100644 --- a/tests/test_litellm/proxy/management_endpoints/scim/test_scim_v2_endpoints.py +++ b/tests/test_litellm/proxy/management_endpoints/scim/test_scim_v2_endpoints.py @@ -7,7 +7,8 @@ from typing import Final from unittest.mock import AsyncMock, MagicMock, call import pytest -from fastapi import HTTPException +from fastapi import FastAPI, HTTPException +from httpx import ASGITransport, AsyncClient from pytest_mock import MockerFixture from litellm.proxy._types import ( @@ -31,6 +32,7 @@ from litellm.proxy.management_endpoints.scim.scim_v2 import ( _handle_group_membership_changes, _handle_team_membership_changes, _parse_member_entries, + _premium_user_check, _process_group_patch_operations, _recompute_scim_member_roles, _resolve_group_member_ids, @@ -45,8 +47,10 @@ from litellm.proxy.management_endpoints.scim.scim_v2 import ( patch_group, patch_team_membership, patch_user, + scim_router, update_group, update_user, + user_api_key_auth, ) from litellm.types.proxy.management_endpoints.scim_v2 import ( SCIM_ENTERPRISE_USER_SCHEMA, @@ -484,6 +488,48 @@ async def test_scim_create_user_respects_default_role_set_via_ui(mocker, monkeyp ) +@pytest.fixture +def scim_test_client(): + """An in-process SCIM application with authorization dependencies bypassed.""" + app = FastAPI() + app.dependency_overrides[_premium_user_check] = lambda: None + app.dependency_overrides[user_api_key_auth] = lambda: UserAPIKeyAuth(user_role=LitellmUserRoles.PROXY_ADMIN) + app.include_router(scim_router) + return AsyncClient(transport=ASGITransport(app=app), base_url="http://test") + + +@pytest.mark.asyncio +@pytest.mark.parametrize("endpoint", ["Users", "Groups"]) +@pytest.mark.parametrize(("requested_count", "effective_count"), [(0, 0), (200, 100), (1000, 100)]) +async def test_scim_collection_endpoints_clamp_requested_page_size( + scim_test_client, endpoint, requested_count, effective_count, mocker +): + """SCIM list endpoints accept zero and cap larger client page requests.""" + mock_prisma_client = MagicMock() + mock_prisma_client.db = MagicMock() + table = MagicMock() + table.find_many = AsyncMock(return_value=[]) + table.count = AsyncMock(return_value=0) + mock_prisma_client.db.litellm_usertable = table + mock_prisma_client.db.litellm_teamtable = table + mocker.patch( # test-quality-ok: HTTP validation requires an in-memory database boundary. + "litellm.proxy.management_endpoints.scim.scim_v2._get_prisma_client_or_raise_exception", + AsyncMock(return_value=mock_prisma_client), + ) + + async with scim_test_client as client: + response = await client.get(f"/scim/v2/{endpoint}?startIndex=1&count={requested_count}") + + assert response.status_code == 200 + table.find_many.assert_awaited_once_with( + where={}, + skip=0, + take=effective_count, + order={"created_at": "desc"}, + ) + assert response.json()["itemsPerPage"] == 0 + + @pytest.mark.asyncio async def test_get_users_filters_username_by_exposed_scim_username_for_okta(mocker): """ diff --git a/tests/test_litellm/proxy/pass_through_endpoints/llm_provider_handlers/test_typesafe_passthrough_logging_handler.py b/tests/test_litellm/proxy/pass_through_endpoints/llm_provider_handlers/test_typesafe_passthrough_logging_handler.py new file mode 100644 index 00000000000..345eeeedc31 --- /dev/null +++ b/tests/test_litellm/proxy/pass_through_endpoints/llm_provider_handlers/test_typesafe_passthrough_logging_handler.py @@ -0,0 +1,134 @@ +from datetime import datetime +from unittest.mock import MagicMock + +import httpx +import pytest + +import litellm +from litellm.proxy.pass_through_endpoints.llm_provider_handlers.typesafe_passthrough_logging_handler import ( + TypeSafePassthroughLoggingHandler, +) +from litellm.proxy.pass_through_endpoints.success_handler import PassThroughEndpointLogging + + +@pytest.fixture(autouse=True) +def local_model_cost_map(monkeypatch: pytest.MonkeyPatch): + monkeypatch.setenv("LITELLM_LOCAL_MODEL_COST_MAP", "True") + monkeypatch.setattr(litellm, "model_cost", litellm.get_model_cost_map(url="")) + + +def _response() -> httpx.Response: + return httpx.Response( + 200, + request=httpx.Request("POST", "https://api.typesafe.ai/v1/systemone"), + json={"model": "jev-1.13.0"}, + ) + + +def _logging_obj() -> MagicMock: + logging_obj = MagicMock() + logging_obj.model_call_details = {} + return logging_obj + + +def _handler_result(response_body: dict, request_body: dict) -> dict: + return TypeSafePassthroughLoggingHandler.typesafe_passthrough_handler( + httpx_response=_response(), + response_body=response_body, + logging_obj=_logging_obj(), + url_route="https://api.typesafe.ai/v1/systemone", + result='{"answers": {}}', + start_time=datetime.now(), + end_time=datetime.now(), + cache_hit=False, + request_body=request_body, + ) + + +def test_uses_registry_pricing_and_standard_usage(): + logging_obj = _logging_obj() + model_key = "typesafe/jev-1.13.0" + model_cost = litellm.model_cost[model_key] + response = TypeSafePassthroughLoggingHandler.typesafe_passthrough_handler( + httpx_response=_response(), + response_body={"model": "jev-1.13.0", "usage": {"input_tokens": 312, "output_tokens": 48}}, + logging_obj=logging_obj, + url_route="https://api.typesafe.ai/v1/systemone", + result='{"answers": {}}', + start_time=datetime.now(), + end_time=datetime.now(), + cache_hit=False, + request_body={"model": "jev-latest"}, + ) + + expected_cost = 312 * model_cost["input_cost_per_token"] + 48 * model_cost["output_cost_per_token"] + assert response["kwargs"]["response_cost"] == pytest.approx(expected_cost) + assert response["kwargs"]["combined_usage_object"].prompt_tokens == 312 + assert response["kwargs"]["combined_usage_object"].completion_tokens == 48 + assert response["kwargs"]["combined_usage_object"].total_tokens == 360 + + +def test_falls_back_to_request_model_when_response_model_is_missing(): + result = _handler_result( + {"usage": {"input_tokens": 10, "output_tokens": 2}}, + {"model": "jev-latest"}, + ) + + model_cost = litellm.model_cost["typesafe/jev-latest"] + expected_cost = 10 * model_cost["input_cost_per_token"] + 2 * model_cost["output_cost_per_token"] + assert result["kwargs"]["model"] == "typesafe/jev-latest" + assert result["kwargs"]["response_cost"] == pytest.approx(expected_cost) + + +def test_call_naming_no_model_is_logged_as_unknown_and_never_priced_as_a_registry_model(): + result = _handler_result({"usage": {"input_tokens": 10, "output_tokens": 2}}, {}) + + assert result["kwargs"]["model"] == "typesafe/unknown" + assert result["kwargs"]["response_cost"] == 0.0 + + +def test_missing_usage_is_zero_cost(): + result = _handler_result({"model": "jev-1.13.0"}, {"model": "jev-latest"}) + + assert result["kwargs"]["response_cost"] == 0.0 + + +def test_records_model_provider_and_cost_on_logging_details(): + logging_obj = _logging_obj() + result = TypeSafePassthroughLoggingHandler.typesafe_passthrough_handler( + httpx_response=_response(), + response_body={"model": "jev-1.13.0", "usage": {"input_tokens": 1, "output_tokens": 0}}, + logging_obj=logging_obj, + url_route="https://api.typesafe.ai/v1/systemone", + result="{}", + start_time=datetime.now(), + end_time=datetime.now(), + cache_hit=False, + request_body={"model": "jev-latest"}, + ) + + assert result["kwargs"]["model"] == "typesafe/jev-1.13.0" + assert result["kwargs"]["custom_llm_provider"] == "typesafe" + assert result["kwargs"]["response_cost"] > 0 + assert logging_obj.model_call_details["model"] == "typesafe/jev-1.13.0" + assert logging_obj.model_call_details["custom_llm_provider"] == "typesafe" + assert logging_obj.model_call_details["response_cost"] == result["kwargs"]["response_cost"] + + +def test_success_handler_dispatches_to_typesafe_handler(): + logging_obj = _logging_obj() + normalized = PassThroughEndpointLogging().normalize_llm_passthrough_logging_payload( + httpx_response=_response(), + response_body={"model": "jev-1.13.0", "usage": {"input_tokens": 1, "output_tokens": 0}}, + request_body={"model": "jev-latest"}, + logging_obj=logging_obj, + url_route="https://api.typesafe.ai/v1/systemone", + result="{}", + start_time=datetime.now(), + end_time=datetime.now(), + cache_hit=False, + custom_llm_provider="typesafe", + ) + + assert normalized["kwargs"]["custom_llm_provider"] == "typesafe" + assert normalized["kwargs"]["model"] == "typesafe/jev-1.13.0" diff --git a/tests/test_litellm/proxy/pass_through_endpoints/test_llm_pass_through_endpoints.py b/tests/test_litellm/proxy/pass_through_endpoints/test_llm_pass_through_endpoints.py index 6e82c90514d..901c5318442 100644 --- a/tests/test_litellm/proxy/pass_through_endpoints/test_llm_pass_through_endpoints.py +++ b/tests/test_litellm/proxy/pass_through_endpoints/test_llm_pass_through_endpoints.py @@ -9,6 +9,7 @@ from types import MappingProxyType, SimpleNamespace from typing import Final from unittest import mock from unittest.mock import AsyncMock, MagicMock, Mock, patch +from urllib.parse import parse_qs import httpx import pytest @@ -43,6 +44,7 @@ from litellm.proxy.pass_through_endpoints.llm_passthrough_endpoints import ( mistral_proxy_route, relay_nvidia_nim_request, openai_proxy_route, + typesafe_proxy_route, vertex_discovery_proxy_route, vertex_proxy_route, vllm_proxy_route, @@ -6136,3 +6138,51 @@ class TestAzureRelayDeploymentSegment: ) assert [call["model"] for call in captured] == ["gpt", "gpt"] + + +class TestTypeSafePassthroughRoute: + @staticmethod + def _request(body: object, query_params: Mapping[str, str] | None = None) -> MagicMock: + request = MagicMock(spec=Request) + request.method = "POST" + request.query_params = query_params or {} + request.json = AsyncMock(return_value=body) + return request + + @pytest.mark.asyncio + async def test_forwards_target_auth_headers_provider_and_query(self, monkeypatch): + monkeypatch.setenv("TYPESAFE_API_KEY", "typesafe-test-key") + monkeypatch.setenv("TYPESAFE_API_BASE", "https://typesafe.example/base") + + async def fake_upstream(request, *_args): + target: Final = create_route.call_args.kwargs["target"] + upstream_url: Final = httpx.URL(target).copy_merge_params(request.query_params) + return {"upstream_query": parse_qs(upstream_url.query.decode())} + + endpoint_func = AsyncMock(side_effect=fake_upstream) + create_route = Mock(return_value=endpoint_func) + monkeypatch.setattr( + "litellm.proxy.pass_through_endpoints.llm_passthrough_endpoints.create_pass_through_route", + create_route, + ) + + request = self._request({"state": "x"}, {"trace": "yes"}) + result = await typesafe_proxy_route( + endpoint="v1/systemone", + request=request, + fastapi_response=MagicMock(spec=Response), + user_api_key_dict=UserAPIKeyAuth(api_key="virtual-key"), + ) + + assert result == {"upstream_query": {"trace": ["yes"]}} + endpoint_func.assert_awaited_once() + create_route.assert_called_once_with( + endpoint="v1/systemone", + target="https://typesafe.example/base/v1/systemone", + custom_headers={ + "Authorization": "Bearer typesafe-test-key", + "Content-Type": "application/json", + }, + custom_llm_provider="typesafe", + is_streaming_request=False, + ) diff --git a/tests/test_litellm/proxy/spend_tracking/test_spend_management_endpoints.py b/tests/test_litellm/proxy/spend_tracking/test_spend_management_endpoints.py index 772c5f674d5..8d15fb094d5 100644 --- a/tests/test_litellm/proxy/spend_tracking/test_spend_management_endpoints.py +++ b/tests/test_litellm/proxy/spend_tracking/test_spend_management_endpoints.py @@ -3745,7 +3745,7 @@ class TestSpendLogsPayload: "model": "gpt-4o", "user": "", "team_id": "", - "metadata": '{"applied_guardrails": [], "attempted_fallbacks": null, "original_model_group": null, "batch_models": null, "batch_successful_requests": null, "batch_failed_requests": null, "mcp_tool_call_metadata": null, "vector_store_request_metadata": null, "routing_decision": null, "internal_call_origin": null, "guardrail_information": null, "compression_savings": null, "litellm_gateway_injected_cache": null, "router_metadata": null, "usage_object": {"completion_tokens": 20, "prompt_tokens": 10, "total_tokens": 30, "completion_tokens_details": null, "prompt_tokens_details": null}, "model_map_information": {"model_map_key": "gpt-4o", "model_map_value": {"key": "gpt-4o", "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, "input_cost_per_token": 2.5e-06, "cache_creation_input_token_cost": null, "cache_read_input_token_cost": 1.25e-06, "input_cost_per_character": null, "input_cost_per_token_above_128k_tokens": null, "input_cost_per_token_above_200k_tokens": null, "input_cost_per_query": null, "input_cost_per_second": null, "input_cost_per_audio_token": null, "input_cost_per_token_batches": 1.25e-06, "output_cost_per_token_batches": 5e-06, "output_cost_per_token": 1e-05, "output_cost_per_audio_token": null, "output_cost_per_character": null, "output_cost_per_token_above_128k_tokens": null, "output_cost_per_character_above_128k_tokens": null, "output_cost_per_token_above_200k_tokens": null, "output_cost_per_second": null, "output_cost_per_reasoning_token": null, "output_cost_per_image": null, "output_vector_size": null, "litellm_provider": "openai", "mode": "chat", "supports_system_messages": true, "supports_response_schema": true, "supports_vision": true, "supports_function_calling": true, "supports_tool_choice": true, "supports_assistant_prefill": false, "supports_prompt_caching": true, "supports_audio_input": false, "supports_audio_output": false, "supports_pdf_input": false, "supports_embedding_image_input": false, "supports_native_streaming": null, "supports_web_search": true, "supports_reasoning": false, "search_context_cost_per_query": {"search_context_size_low": 0.03, "search_context_size_medium": 0.035, "search_context_size_high": 0.05}, "tpm": null, "rpm": null, "supported_openai_params": ["frequency_penalty", "logit_bias", "logprobs", "top_logprobs", "max_tokens", "max_completion_tokens", "modalities", "prediction", "n", "presence_penalty", "seed", "stop", "stream", "stream_options", "temperature", "top_p", "tools", "tool_choice", "function_call", "functions", "max_retries", "extra_headers", "parallel_tool_calls", "audio", "response_format", "user"]}}, "additional_usage_values": {"completion_tokens_details": null, "prompt_tokens_details": null}}', + "metadata": '{"applied_guardrails": [], "attempted_fallbacks": null, "original_model_group": null, "batch_models": null, "batch_successful_requests": null, "batch_failed_requests": null, "mcp_tool_call_metadata": null, "vector_store_request_metadata": null, "routing_decision": null, "internal_call_origin": null, "guardrail_information": null, "compression_savings": null, "litellm_gateway_injected_cache": null, "router_metadata": null, "azure_spillover": null, "usage_object": {"completion_tokens": 20, "prompt_tokens": 10, "total_tokens": 30, "completion_tokens_details": null, "prompt_tokens_details": null}, "model_map_information": {"model_map_key": "gpt-4o", "model_map_value": {"key": "gpt-4o", "max_tokens": 16384, "max_input_tokens": 128000, "max_output_tokens": 16384, "input_cost_per_token": 2.5e-06, "cache_creation_input_token_cost": null, "cache_read_input_token_cost": 1.25e-06, "input_cost_per_character": null, "input_cost_per_token_above_128k_tokens": null, "input_cost_per_token_above_200k_tokens": null, "input_cost_per_query": null, "input_cost_per_second": null, "input_cost_per_audio_token": null, "input_cost_per_token_batches": 1.25e-06, "output_cost_per_token_batches": 5e-06, "output_cost_per_token": 1e-05, "output_cost_per_audio_token": null, "output_cost_per_character": null, "output_cost_per_token_above_128k_tokens": null, "output_cost_per_character_above_128k_tokens": null, "output_cost_per_token_above_200k_tokens": null, "output_cost_per_second": null, "output_cost_per_reasoning_token": null, "output_cost_per_image": null, "output_vector_size": null, "litellm_provider": "openai", "mode": "chat", "supports_system_messages": true, "supports_response_schema": true, "supports_vision": true, "supports_function_calling": true, "supports_tool_choice": true, "supports_assistant_prefill": false, "supports_prompt_caching": true, "supports_audio_input": false, "supports_audio_output": false, "supports_pdf_input": false, "supports_embedding_image_input": false, "supports_native_streaming": null, "supports_web_search": true, "supports_reasoning": false, "search_context_cost_per_query": {"search_context_size_low": 0.03, "search_context_size_medium": 0.035, "search_context_size_high": 0.05}, "tpm": null, "rpm": null, "supported_openai_params": ["frequency_penalty", "logit_bias", "logprobs", "top_logprobs", "max_tokens", "max_completion_tokens", "modalities", "prediction", "n", "presence_penalty", "seed", "stop", "stream", "stream_options", "temperature", "top_p", "tools", "tool_choice", "function_call", "functions", "max_retries", "extra_headers", "parallel_tool_calls", "audio", "response_format", "user"]}}, "additional_usage_values": {"completion_tokens_details": null, "prompt_tokens_details": null}}', "cache_key": "Cache OFF", "spend": 0.00022500000000000002, "total_tokens": 30, @@ -3841,7 +3841,7 @@ class TestSpendLogsPayload: "model": "claude-4-sonnet-20250514", "user": "", "team_id": "", - "metadata": '{"applied_guardrails": [], "attempted_fallbacks": null, "original_model_group": null, "batch_models": null, "batch_successful_requests": null, "batch_failed_requests": null, "mcp_tool_call_metadata": null, "vector_store_request_metadata": null, "routing_decision": null, "internal_call_origin": null, "guardrail_information": null, "compression_savings": null, "litellm_gateway_injected_cache": null, "router_metadata": null, "usage_object": {"completion_tokens": 503, "prompt_tokens": 2095, "total_tokens": 2598, "completion_tokens_details": null, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 0}, "cache_creation_input_tokens": 0, "cache_read_input_tokens": 0}, "model_map_information": {"model_map_key": "claude-4-sonnet-20250514", "model_map_value": {"key": "claude-4-sonnet-20250514", "max_tokens": 128000, "max_input_tokens": 200000, "max_output_tokens": 128000, "input_cost_per_token": 3e-06, "cache_creation_input_token_cost": 3.75e-06, "cache_read_input_token_cost": 3e-07, "input_cost_per_character": null, "input_cost_per_token_above_128k_tokens": null, "input_cost_per_token_above_200k_tokens": null, "input_cost_per_query": null, "input_cost_per_second": null, "input_cost_per_audio_token": null, "input_cost_per_token_batches": null, "output_cost_per_token_batches": null, "output_cost_per_token": 1.5e-05, "output_cost_per_audio_token": null, "output_cost_per_character": null, "output_cost_per_token_above_128k_tokens": null, "output_cost_per_character_above_128k_tokens": null, "output_cost_per_token_above_200k_tokens": null, "output_cost_per_second": null, "output_cost_per_image": null, "output_vector_size": null, "litellm_provider": "anthropic", "mode": "chat", "supports_system_messages": null, "supports_response_schema": true, "supports_vision": true, "supports_function_calling": true, "supports_tool_choice": true, "supports_assistant_prefill": true, "supports_prompt_caching": true, "supports_audio_input": false, "supports_audio_output": false, "supports_pdf_input": true, "supports_embedding_image_input": false, "supports_native_streaming": null, "supports_web_search": false, "supports_reasoning": true, "search_context_cost_per_query": null, "tpm": null, "rpm": null, "supported_openai_params": ["stream", "stop", "temperature", "top_p", "max_tokens", "max_completion_tokens", "tools", "tool_choice", "extra_headers", "parallel_tool_calls", "response_format", "user", "reasoning_effort", "thinking"]}}, "additional_usage_values": {"completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": null, "rejected_prediction_tokens": null, "text_tokens": 503, "image_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 0, "text_tokens": null, "image_tokens": null}, "cache_creation_input_tokens": 0, "cache_read_input_tokens": 0}}', + "metadata": '{"applied_guardrails": [], "attempted_fallbacks": null, "original_model_group": null, "batch_models": null, "batch_successful_requests": null, "batch_failed_requests": null, "mcp_tool_call_metadata": null, "vector_store_request_metadata": null, "routing_decision": null, "internal_call_origin": null, "guardrail_information": null, "compression_savings": null, "litellm_gateway_injected_cache": null, "router_metadata": null, "azure_spillover": null, "usage_object": {"completion_tokens": 503, "prompt_tokens": 2095, "total_tokens": 2598, "completion_tokens_details": null, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 0}, "cache_creation_input_tokens": 0, "cache_read_input_tokens": 0}, "model_map_information": {"model_map_key": "claude-4-sonnet-20250514", "model_map_value": {"key": "claude-4-sonnet-20250514", "max_tokens": 128000, "max_input_tokens": 200000, "max_output_tokens": 128000, "input_cost_per_token": 3e-06, "cache_creation_input_token_cost": 3.75e-06, "cache_read_input_token_cost": 3e-07, "input_cost_per_character": null, "input_cost_per_token_above_128k_tokens": null, "input_cost_per_token_above_200k_tokens": null, "input_cost_per_query": null, "input_cost_per_second": null, "input_cost_per_audio_token": null, "input_cost_per_token_batches": null, "output_cost_per_token_batches": null, "output_cost_per_token": 1.5e-05, "output_cost_per_audio_token": null, "output_cost_per_character": null, "output_cost_per_token_above_128k_tokens": null, "output_cost_per_character_above_128k_tokens": null, "output_cost_per_token_above_200k_tokens": null, "output_cost_per_second": null, "output_cost_per_image": null, "output_vector_size": null, "litellm_provider": "anthropic", "mode": "chat", "supports_system_messages": null, "supports_response_schema": true, "supports_vision": true, "supports_function_calling": true, "supports_tool_choice": true, "supports_assistant_prefill": true, "supports_prompt_caching": true, "supports_audio_input": false, "supports_audio_output": false, "supports_pdf_input": true, "supports_embedding_image_input": false, "supports_native_streaming": null, "supports_web_search": false, "supports_reasoning": true, "search_context_cost_per_query": null, "tpm": null, "rpm": null, "supported_openai_params": ["stream", "stop", "temperature", "top_p", "max_tokens", "max_completion_tokens", "tools", "tool_choice", "extra_headers", "parallel_tool_calls", "response_format", "user", "reasoning_effort", "thinking"]}}, "additional_usage_values": {"completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": null, "rejected_prediction_tokens": null, "text_tokens": 503, "image_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 0, "text_tokens": null, "image_tokens": null}, "cache_creation_input_tokens": 0, "cache_read_input_tokens": 0}}', "cache_key": "Cache OFF", "spend": 0.01383, "total_tokens": 2598, @@ -3935,7 +3935,7 @@ class TestSpendLogsPayload: "model": "claude-4-sonnet-20250514", "user": "", "team_id": "", - "metadata": '{"applied_guardrails": [], "attempted_fallbacks": 0, "original_model_group": "my-anthropic-model-group", "batch_models": null, "batch_successful_requests": null, "batch_failed_requests": null, "mcp_tool_call_metadata": null, "vector_store_request_metadata": null, "routing_decision": null, "internal_call_origin": null, "guardrail_information": null, "compression_savings": null, "litellm_gateway_injected_cache": null, "router_metadata": null, "usage_object": {"completion_tokens": 503, "prompt_tokens": 2095, "total_tokens": 2598, "completion_tokens_details": null, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 0}, "cache_creation_input_tokens": 0, "cache_read_input_tokens": 0}, "model_map_information": {"model_map_key": "claude-4-sonnet-20250514", "model_map_value": {"key": "claude-4-sonnet-20250514", "max_tokens": 128000, "max_input_tokens": 200000, "max_output_tokens": 128000, "input_cost_per_token": 3e-06, "cache_creation_input_token_cost": 3.75e-06, "cache_read_input_token_cost": 3e-07, "input_cost_per_character": null, "input_cost_per_token_above_128k_tokens": null, "input_cost_per_token_above_200k_tokens": null, "input_cost_per_query": null, "input_cost_per_second": null, "input_cost_per_audio_token": null, "input_cost_per_token_batches": null, "output_cost_per_token_batches": null, "output_cost_per_token": 1.5e-05, "output_cost_per_audio_token": null, "output_cost_per_character": null, "output_cost_per_token_above_128k_tokens": null, "output_cost_per_character_above_128k_tokens": null, "output_cost_per_token_above_200k_tokens": null, "output_cost_per_second": null, "output_cost_per_image": null, "output_vector_size": null, "litellm_provider": "anthropic", "mode": "chat", "supports_system_messages": null, "supports_response_schema": true, "supports_vision": true, "supports_function_calling": true, "supports_tool_choice": true, "supports_assistant_prefill": true, "supports_prompt_caching": true, "supports_audio_input": false, "supports_audio_output": false, "supports_pdf_input": true, "supports_embedding_image_input": false, "supports_native_streaming": null, "supports_web_search": false, "supports_reasoning": true, "search_context_cost_per_query": null, "tpm": null, "rpm": null, "supported_openai_params": ["stream", "stop", "temperature", "top_p", "max_tokens", "max_completion_tokens", "tools", "tool_choice", "extra_headers", "parallel_tool_calls", "response_format", "user", "reasoning_effort", "thinking"]}}, "additional_usage_values": {"completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": null, "rejected_prediction_tokens": null, "text_tokens": 503, "image_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 0, "text_tokens": null, "image_tokens": null}, "cache_creation_input_tokens": 0, "cache_read_input_tokens": 0}}', + "metadata": '{"applied_guardrails": [], "attempted_fallbacks": 0, "original_model_group": "my-anthropic-model-group", "batch_models": null, "batch_successful_requests": null, "batch_failed_requests": null, "mcp_tool_call_metadata": null, "vector_store_request_metadata": null, "routing_decision": null, "internal_call_origin": null, "guardrail_information": null, "compression_savings": null, "litellm_gateway_injected_cache": null, "router_metadata": null, "azure_spillover": null, "usage_object": {"completion_tokens": 503, "prompt_tokens": 2095, "total_tokens": 2598, "completion_tokens_details": null, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 0}, "cache_creation_input_tokens": 0, "cache_read_input_tokens": 0}, "model_map_information": {"model_map_key": "claude-4-sonnet-20250514", "model_map_value": {"key": "claude-4-sonnet-20250514", "max_tokens": 128000, "max_input_tokens": 200000, "max_output_tokens": 128000, "input_cost_per_token": 3e-06, "cache_creation_input_token_cost": 3.75e-06, "cache_read_input_token_cost": 3e-07, "input_cost_per_character": null, "input_cost_per_token_above_128k_tokens": null, "input_cost_per_token_above_200k_tokens": null, "input_cost_per_query": null, "input_cost_per_second": null, "input_cost_per_audio_token": null, "input_cost_per_token_batches": null, "output_cost_per_token_batches": null, "output_cost_per_token": 1.5e-05, "output_cost_per_audio_token": null, "output_cost_per_character": null, "output_cost_per_token_above_128k_tokens": null, "output_cost_per_character_above_128k_tokens": null, "output_cost_per_token_above_200k_tokens": null, "output_cost_per_second": null, "output_cost_per_image": null, "output_vector_size": null, "litellm_provider": "anthropic", "mode": "chat", "supports_system_messages": null, "supports_response_schema": true, "supports_vision": true, "supports_function_calling": true, "supports_tool_choice": true, "supports_assistant_prefill": true, "supports_prompt_caching": true, "supports_audio_input": false, "supports_audio_output": false, "supports_pdf_input": true, "supports_embedding_image_input": false, "supports_native_streaming": null, "supports_web_search": false, "supports_reasoning": true, "search_context_cost_per_query": null, "tpm": null, "rpm": null, "supported_openai_params": ["stream", "stop", "temperature", "top_p", "max_tokens", "max_completion_tokens", "tools", "tool_choice", "extra_headers", "parallel_tool_calls", "response_format", "user", "reasoning_effort", "thinking"]}}, "additional_usage_values": {"completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": null, "rejected_prediction_tokens": null, "text_tokens": 503, "image_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 0, "text_tokens": null, "image_tokens": null}, "cache_creation_input_tokens": 0, "cache_read_input_tokens": 0}}', "cache_key": "Cache OFF", "spend": 0.01383, "total_tokens": 2598, diff --git a/tests/test_litellm/proxy/spend_tracking/test_spend_tracking_utils.py b/tests/test_litellm/proxy/spend_tracking/test_spend_tracking_utils.py index 7663bd83790..f2273924b30 100644 --- a/tests/test_litellm/proxy/spend_tracking/test_spend_tracking_utils.py +++ b/tests/test_litellm/proxy/spend_tracking/test_spend_tracking_utils.py @@ -4850,3 +4850,58 @@ def test_spend_log_request_id_is_the_response_id_a_bridged_messages_caller_recei ) == "resp_01Lit6806Bridged" ) + + +def test_azure_spillover_stamped_from_response_headers(): + """Raw provider response headers on the logging kwargs mark the request as spilled.""" + kwargs: Final = { + **_routed_call_kwargs({"id": "mi-1"}), + "response_headers": { + "x-ms-is-spilled-over": "true", + "x-ms-spillover-from-deployment": "my-ptu", + }, + } + payload = get_logging_payload( + kwargs=kwargs, + response_obj=litellm.ModelResponse(id="chatcmpl-spill-raw", choices=[], usage=litellm.Usage()), + start_time=datetime.datetime.now(timezone.utc), + end_time=datetime.datetime.now(timezone.utc), + ) + metadata = json.loads(payload["metadata"]) + assert metadata["azure_spillover"] == {"from_deployment": "my-ptu"} + + +def test_azure_spillover_stamped_from_standard_logging_additional_headers(): + """Streaming requests carry the processed llm_provider- headers on the standard payload.""" + kwargs: Final = { + **_routed_call_kwargs({"id": "mi-1"}), + "standard_logging_object": { + "hidden_params": { + "additional_headers": { + "llm_provider-x-ms-is-spilled-over": "true", + "llm_provider-x-ms-spillover-from-deployment": "my-ptu", + } + }, + "metadata": {}, + "model_map_information": None, + }, + } + payload = get_logging_payload( + kwargs=kwargs, + response_obj=litellm.ModelResponse(id="chatcmpl-spill-sl", choices=[], usage=litellm.Usage()), + start_time=datetime.datetime.now(timezone.utc), + end_time=datetime.datetime.now(timezone.utc), + ) + metadata = json.loads(payload["metadata"]) + assert metadata["azure_spillover"] == {"from_deployment": "my-ptu"} + + +def test_azure_spillover_absent_without_spillover_headers(): + payload = get_logging_payload( + kwargs=_routed_call_kwargs({"id": "mi-1"}), + response_obj=litellm.ModelResponse(id="chatcmpl-no-spill", choices=[], usage=litellm.Usage()), + start_time=datetime.datetime.now(timezone.utc), + end_time=datetime.datetime.now(timezone.utc), + ) + metadata = json.loads(payload["metadata"]) + assert metadata["azure_spillover"] is None diff --git a/tests/test_litellm/responses/litellm_completion_transformation/test_handler.py b/tests/test_litellm/responses/litellm_completion_transformation/test_handler.py index b78dabbfe48..bb374b90f4e 100644 --- a/tests/test_litellm/responses/litellm_completion_transformation/test_handler.py +++ b/tests/test_litellm/responses/litellm_completion_transformation/test_handler.py @@ -12,14 +12,21 @@ capture the forwarded kwargs; if the flag-setting line is removed the captured kwargs lack the flag and these tests fail. """ +import json +from collections.abc import Mapping +from typing import Final from unittest.mock import patch +import httpx import pytest - +import litellm +from litellm.llms.custom_httpx.http_handler import AsyncHTTPHandler from litellm.responses.litellm_completion_transformation.handler import ( LiteLLMCompletionTransformationHandler, ) +from litellm.types.llms.openai import ResponsesAPIResponse +from litellm.types.utils import ADDRESSED_RESPONSE_ID_FIELD class _StopForwarding(Exception): @@ -170,3 +177,49 @@ async def test_async_fallback_returns_hoisted_nested_custom_tool_call_as_custom_ tool_calls = [(item.type, item.name, item.input) for item in response.output if item.type == "custom_tool_call"] assert tool_calls == [("custom_tool_call", "exec", "ls")] + + +class _RecordingAnthropicHandler: + def __init__(self, reply: Mapping[str, object]) -> None: + self.reply: Final = reply + self.request_body: Mapping[str, object] | None = None + + def __call__(self, request: httpx.Request) -> httpx.Response: + self.request_body = json.loads(request.content) + return httpx.Response(200, json=dict(self.reply), request=request) + + +_ANTHROPIC_MESSAGE_PAYLOAD: Final = { + "id": "msg_turn_two", + "type": "message", + "role": "assistant", + "model": "claude-sonnet-4-6", + "content": [{"type": "text", "text": "14"}], + "stop_reason": "end_turn", + "stop_sequence": None, + "usage": {"input_tokens": 12, "output_tokens": 1}, +} + + +@pytest.mark.asyncio +async def test_bridged_follow_up_turn_keeps_the_addressed_response_id_off_the_provider_body(): + provider: Final = _RecordingAnthropicHandler(_ANTHROPIC_MESSAGE_PAYLOAD) + client: Final = AsyncHTTPHandler() + client.client = httpx.AsyncClient(transport=httpx.MockTransport(provider)) + + response = await litellm.aresponses( + model="azure_ai/claude-sonnet-4-6", + api_base="https://fake-foundry-resource.services.ai.azure.com", + api_key="fake-api-key", + input="Double it", + previous_response_id="resp_turn_one", + client=client, + **{ADDRESSED_RESPONSE_ID_FIELD: "resp_turn_one"}, + ) + + assert provider.request_body is not None, "the bridged turn never reached the provider" + assert ADDRESSED_RESPONSE_ID_FIELD not in provider.request_body, ( + f"the addressed response id reached the provider body: {sorted(provider.request_body)}" + ) + assert isinstance(response, ResponsesAPIResponse) + assert [item.type for item in response.output] == ["message"] diff --git a/tests/test_litellm/test_typesafe_model_metadata.py b/tests/test_litellm/test_typesafe_model_metadata.py new file mode 100644 index 00000000000..a27180afbe9 --- /dev/null +++ b/tests/test_litellm/test_typesafe_model_metadata.py @@ -0,0 +1,17 @@ +import pytest + +import litellm + + +@pytest.fixture(autouse=True) +def local_model_cost_map(monkeypatch: pytest.MonkeyPatch): + monkeypatch.setenv("LITELLM_LOCAL_MODEL_COST_MAP", "True") + monkeypatch.setattr(litellm, "model_cost", litellm.get_model_cost_map(url="")) + + +def test_typesafe_models_share_pricing_and_provider_metadata(): + entries = [litellm.model_cost[f"typesafe/{model}"] for model in ("jev-1.13.0", "jev-latest", "jev-preview")] + + assert {entry["input_cost_per_token"] for entry in entries} == {entries[0]["input_cost_per_token"]} + assert {entry["output_cost_per_token"] for entry in entries} == {entries[0]["output_cost_per_token"]} + assert {entry["litellm_provider"] for entry in entries} == {"typesafe"} diff --git a/tests/test_litellm/test_utils.py b/tests/test_litellm/test_utils.py index 0d5d507101a..4618c156046 100644 --- a/tests/test_litellm/test_utils.py +++ b/tests/test_litellm/test_utils.py @@ -46,6 +46,7 @@ from litellm.types.utils import ( PromptTokensDetailsWrapper, StreamingChoices, Usage, + ADDRESSED_RESPONSE_ID_FIELD, all_litellm_params, bedrock_batch_litellm_params, ) @@ -818,6 +819,7 @@ def test_aaamodel_prices_and_context_window_json_is_valid(): "container", "image_edit", "embedding", + "evaluation", "guardrail", "image_generation", "video_generation", @@ -3546,7 +3548,7 @@ _FIREWORKS_MODELS = [ "accounts/fireworks/models/minimax-m3", 512000, 512000, - None, + True, True, ), ( @@ -3654,8 +3656,7 @@ def _assert_fireworks_entry( assert info["supports_tool_choice"] is True assert info["supports_reasoning"] is expected_reasoning assert info["supports_response_schema"] is True - if expected_vision is not None: - assert info["supports_vision"] is expected_vision + assert info["supports_vision"] is expected_vision @pytest.fixture @@ -4787,6 +4788,20 @@ def test_get_litellm_params_keys_never_reach_the_provider(): ) +def test_addressed_response_id_never_reaches_the_provider(): + kwargs = { + "a_real_provider_specific_param": 1, + ADDRESSED_RESPONSE_ID_FIELD: "resp_addressed-by-the-client", + } + + non_default = get_non_default_completion_params(kwargs) + + assert non_default == {"a_real_provider_specific_param": 1}, ( + "the addressed response id leaked into the provider params: " + f"{sorted(set(non_default) - {'a_real_provider_specific_param'})}" + ) + + def test_bedrock_batch_params_never_reach_the_provider(): """A Bedrock managed-batch deployment carries aws_batch_role_arn / s3_* / bedrock_tags in its litellm_params, and the same deployment also serves chat. diff --git a/ui/litellm-dashboard/src/app/(dashboard)/cost-optimization/_components/CacheLeakageCard.test.tsx b/ui/litellm-dashboard/src/app/(dashboard)/cost-optimization/_components/CacheLeakageCard.test.tsx index 54af13d8a90..a97f23b3334 100644 --- a/ui/litellm-dashboard/src/app/(dashboard)/cost-optimization/_components/CacheLeakageCard.test.tsx +++ b/ui/litellm-dashboard/src/app/(dashboard)/cost-optimization/_components/CacheLeakageCard.test.tsx @@ -123,11 +123,11 @@ describe("CacheLeakageCard", () => { expect(firstDataRow()).toHaveTextContent("alpha"); }); - it("switches to the model view and lists only Anthropic models", () => { + it("switches to the model view and lists models from every provider", () => { renderWith([ dayWithModels("2026-07-12", { "claude-sonnet-5": { prompt_tokens: 5000, cache_read_input_tokens: 0 }, - "gpt-4o": { prompt_tokens: 8000, cache_read_input_tokens: 0 }, + "vertex_ai/gemini-2.5-pro": { prompt_tokens: 8000, cache_read_input_tokens: 2000 }, }), ]); @@ -135,7 +135,7 @@ describe("CacheLeakageCard", () => { expect(screen.getByText("Cache leakage by model")).toBeInTheDocument(); expect(screen.getByText("claude-sonnet-5")).toBeInTheDocument(); - expect(screen.queryByText("gpt-4o")).not.toBeInTheDocument(); + expect(screen.getByText("vertex_ai/gemini-2.5-pro")).toBeInTheDocument(); }); it("shows an empty state when no key used tokens in the range", () => { diff --git a/ui/litellm-dashboard/src/app/(dashboard)/cost-optimization/_components/costOptimizationUtils.test.ts b/ui/litellm-dashboard/src/app/(dashboard)/cost-optimization/_components/costOptimizationUtils.test.ts index 5d2c48e6440..9c3915c812f 100644 --- a/ui/litellm-dashboard/src/app/(dashboard)/cost-optimization/_components/costOptimizationUtils.test.ts +++ b/ui/litellm-dashboard/src/app/(dashboard)/cost-optimization/_components/costOptimizationUtils.test.ts @@ -10,7 +10,6 @@ import { classificationRatePer1kTurns, computeCacheLeakage, formatRangeLabel, - isAnthropicModel, localIsoDay, savingsSeriesOf, toCumulative, @@ -209,20 +208,21 @@ describe("computeCacheLeakage", () => { }); describe("computeCacheLeakage by model", () => { - it("aggregates only Anthropic models and ignores other providers", () => { + it("lists every provider's models, not only Anthropic", () => { const models: Record> = { "claude-sonnet-5": { prompt_tokens: 10000, cache_read_input_tokens: 0 }, - "anthropic/claude-haiku-4-5": { prompt_tokens: 4000, cache_read_input_tokens: 0 }, - "bedrock/anthropic.claude-3-5-sonnet": { prompt_tokens: 2000, cache_read_input_tokens: 0 }, - "gpt-4o": { prompt_tokens: 9000, cache_read_input_tokens: 0 }, - "deepseek-chat": { prompt_tokens: 8000, cache_read_input_tokens: 0 }, + "vertex_ai/gemini-2.5-pro": { prompt_tokens: 9000, cache_read_input_tokens: 3000 }, + "bedrock/openai.gpt-5.6-luna": { prompt_tokens: 8000, cache_read_input_tokens: 0 }, + "deepseek-chat": { prompt_tokens: 4000, cache_read_input_tokens: 0 }, }; const { rows } = computeCacheLeakage([modelDay("2026-07-01", models)], "model"); expect(rows.map((r) => r.id)).toEqual([ "claude-sonnet-5", - "anthropic/claude-haiku-4-5", - "bedrock/anthropic.claude-3-5-sonnet", + "bedrock/openai.gpt-5.6-luna", + "vertex_ai/gemini-2.5-pro", + "deepseek-chat", ]); + expect(rows.find((r) => r.id === "vertex_ai/gemini-2.5-pro")?.cacheHitRatio).toBeCloseTo(1 / 3, 6); }); it("labels model rows by model name with no sublabel", () => { @@ -232,34 +232,20 @@ describe("computeCacheLeakage by model", () => { expect(rows[0].sublabel).toBeNull(); }); - it("prices model leakage at the Anthropic realized cache-read discount", () => { + it("prices model leakage at the realized cache-read discount across providers", () => { const results = [ modelDay("2026-07-01", { "claude-sonnet-5": { prompt_tokens: 1000, cache_read_input_tokens: 1000, prompt_caching_savings_spend: 2.0 }, - "claude-haiku-4-5": { prompt_tokens: 500 }, + "gemini-2.5-flash": { prompt_tokens: 500 }, }), ]; const { rows, netSavingsPerCachedToken } = computeCacheLeakage(results, "model"); expect(netSavingsPerCachedToken).toBeCloseTo(0.002, 6); - expect(rows.map((r) => r.id)).toEqual(["claude-haiku-4-5"]); + expect(rows.map((r) => r.id)).toEqual(["gemini-2.5-flash"]); expect(rows[0].potentialSavings).toBeCloseTo(1.0, 6); }); }); -describe("isAnthropicModel", () => { - it("matches Claude-family models across providers and rejects others", () => { - const anthropic = [ - "claude-sonnet-5", - "anthropic/claude-haiku-4-5", - "bedrock/anthropic.claude-3-5-sonnet", - "vertex_ai/claude-opus-4-8", - ]; - const others = ["gpt-4o", "deepseek-chat", "gemini-2.5-pro", "mistral-large"]; - expect(anthropic.every(isAnthropicModel)).toBe(true); - expect(others.some(isAnthropicModel)).toBe(false); - }); -}); - describe("buildDailyToolSeries", () => { const daily: ToolSpendDailyEntry[] = [ { date: "2026-07-01", tool_name: "search", spend: 1.0, call_count: 1 }, diff --git a/ui/litellm-dashboard/src/app/(dashboard)/cost-optimization/_components/costOptimizationUtils.ts b/ui/litellm-dashboard/src/app/(dashboard)/cost-optimization/_components/costOptimizationUtils.ts index 464c779aa2b..2e6d8208989 100644 --- a/ui/litellm-dashboard/src/app/(dashboard)/cost-optimization/_components/costOptimizationUtils.ts +++ b/ui/litellm-dashboard/src/app/(dashboard)/cost-optimization/_components/costOptimizationUtils.ts @@ -44,8 +44,6 @@ export interface CacheLeakageResult { netSavingsPerCachedToken: number | null; } -export const isAnthropicModel = (model: string): boolean => /claude|anthropic/i.test(model); - interface LeakageAccumulator { alias: string | null; teamId: string | null; @@ -96,7 +94,6 @@ const aggregateByModel = (results: readonly DailyData[]): Map(); for (const day of results) { for (const [model, entry] of Object.entries(day.breakdown?.models ?? {})) { - if (!isAnthropicModel(model)) continue; const acc = byModel.get(model) ?? emptyAccumulator(); byModel.set(model, addMetrics(acc, entry.metrics, null, null)); } diff --git a/ui/litellm-dashboard/src/lib/http/schema.d.ts b/ui/litellm-dashboard/src/lib/http/schema.d.ts index 872875cc535..ba2d36c0d02 100644 --- a/ui/litellm-dashboard/src/lib/http/schema.d.ts +++ b/ui/litellm-dashboard/src/lib/http/schema.d.ts @@ -8544,6 +8544,45 @@ export interface paths { patch?: never; trace?: never; }; + "/management/v1/teams/{team_id}/members/bulk_update": { + parameters: { + query?: never; + header?: never; + path?: never; + cookie?: never; + }; + get?: never; + put?: never; + /** + * Bulk Update Team Member Budgets Action + * @description Set per-member limits for up to 500 members of one team in one call. Same + * authorization and member addressing as `/team/member_update`: proxy admins, the team's + * admins, and admins of the team's organization, with each member named by exactly one of + * `user_id` or `user_email`. Unknown body fields are a 422 and an unknown team is a 404. + * + * Each row is a merge patch of that member's limits: a field left out is untouched, a + * field sent as null is cleared, and clearing the last limit drops the member back to the + * team default. A budget row shared by several memberships, the team default included, is + * copied for the member being patched rather than written in place, so one member's new + * cap never lands on anybody else. + * + * `data` holds one result per requested member, in request order, carrying the limits in + * force after the write. A row is `success: false` with an `error` when it names nobody on + * the team or repeats an earlier row. Roles are not part of this route; `/team/member_update` + * still owns them. + * + * Example curl: + * ``` + * curl --location 'http://0.0.0.0:4000/management/v1/teams/team-1/members/bulk_update' --header 'Authorization: Bearer sk-1234' --header 'Content-Type: application/json' --data '{"members": [{"user_id": "user-1", "max_budget_in_team": 10}, {"user_email": "user-2@example.com", "max_budget_in_team": 10, "budget_duration": "30d"}]}' + * ``` + */ + post: operations["bulk_update_team_member_budgets_action_management_v1_teams__team_id__members_bulk_update_post"]; + delete?: never; + options?: never; + head?: never; + patch?: never; + trace?: never; + }; "/management/v1/users/bulk": { parameters: { query?: never; @@ -16437,6 +16476,30 @@ export interface paths { patch: operations["toolset_mcp_route_toolset__toolset_name__mcp_patch"]; trace?: never; }; + "/typesafe/{endpoint}": { + parameters: { + query?: never; + header?: never; + path?: never; + cookie?: never; + }; + /** + * Typesafe Proxy Route + * @description [Docs](https://docs.litellm.ai/docs/pass_through/typesafe) + */ + get: operations["typesafe_proxy_route_typesafe__endpoint__get"]; + put?: never; + /** + * Typesafe Proxy Route + * @description [Docs](https://docs.litellm.ai/docs/pass_through/typesafe) + */ + post: operations["typesafe_proxy_route_typesafe__endpoint__post"]; + delete?: never; + options?: never; + head?: never; + patch?: never; + trace?: never; + }; "/update/default_team_settings": { parameters: { query?: never; @@ -24928,6 +24991,22 @@ export interface components { [key: string]: unknown; } | null; }; + /** + * BulkTeamMemberBudgetUpdateRequest + * @description Body of `POST /management/v1/teams/{team_id}/members/bulk_update`. + */ + BulkTeamMemberBudgetUpdateRequest: { + /** Members */ + members: components["schemas"]["TeamMemberBudgetPatch"][]; + }; + /** + * BulkTeamMemberBudgetUpdateResponse + * @description `{data: [...]}` with one `TeamMemberBudgetUpdateResult` per requested member, in request order. + */ + BulkTeamMemberBudgetUpdateResponse: { + /** Data */ + data: components["schemas"]["TeamMemberBudgetUpdateResult"][]; + }; /** * BulkTeamMemberDeleteRequest * @description Body of `POST /management/v1/teams/{team_id}/members/bulk_delete`. @@ -37927,6 +38006,57 @@ export interface components { /** User Id */ user_id?: string | null; }; + /** + * TeamMemberBudgetPatch + * @description One member's per-member limits, merge-patch style: a field left out of the row is + * untouched, a field sent as null is cleared, and clearing the last limit drops the + * member back to the team default. + */ + TeamMemberBudgetPatch: { + /** Allowed Models */ + allowed_models?: string[] | null; + /** Budget Duration */ + budget_duration?: string | null; + /** Max Budget In Team */ + max_budget_in_team?: number | null; + /** Rpm Limit */ + rpm_limit?: number | null; + /** Tpm Limit */ + tpm_limit?: number | null; + /** User Email */ + user_email?: string | null; + /** User Id */ + user_id?: string | null; + }; + /** + * TeamMemberBudgetUpdateResult + * @description Outcome for one requested member, in request order, carrying the limits in force + * after the write rather than the ones that were asked for. + */ + TeamMemberBudgetUpdateResult: { + /** Allowed Models */ + allowed_models?: string[] | null; + /** Budget Duration */ + budget_duration?: string | null; + /** Budget Id */ + budget_id?: string | null; + /** Error */ + error?: string | null; + /** Max Budget */ + max_budget?: number | null; + /** Max Budget Source */ + max_budget_source?: ("member" | "team_default") | null; + /** Rpm Limit */ + rpm_limit?: number | null; + /** Success */ + success: boolean; + /** Tpm Limit */ + tpm_limit?: number | null; + /** User Email */ + user_email?: string | null; + /** User Id */ + user_id?: string | null; + }; /** TeamMemberDeleteRequest */ TeamMemberDeleteRequest: { /** Team Id */ @@ -37982,7 +38112,7 @@ export interface components { }; /** * TeamMemberRef - * @description One member to remove, named by exactly one of `user_id` or `user_email`. + * @description One member, named by exactly one of `user_id` or `user_email`. */ TeamMemberRef: { /** User Email */ @@ -52077,6 +52207,44 @@ export interface operations { }; }; }; + bulk_update_team_member_budgets_action_management_v1_teams__team_id__members_bulk_update_post: { + parameters: { + query?: never; + header?: { + /** @description The litellm-changed-by header enables tracking of actions performed by authorized users on behalf of other users, providing an audit trail for accountability */ + "litellm-changed-by"?: string | null; + }; + path: { + team_id: string; + }; + cookie?: never; + }; + requestBody: { + content: { + "application/json": components["schemas"]["BulkTeamMemberBudgetUpdateRequest"]; + }; + }; + responses: { + /** @description Successful Response */ + 200: { + headers: { + [name: string]: unknown; + }; + content: { + "application/json": components["schemas"]["BulkTeamMemberBudgetUpdateResponse"]; + }; + }; + /** @description Validation Error */ + 422: { + headers: { + [name: string]: unknown; + }; + content: { + "application/json": components["schemas"]["HTTPValidationError"]; + }; + }; + }; + }; bulk_create_users_route_management_v1_users_bulk_post: { parameters: { query?: never; @@ -61441,6 +61609,68 @@ export interface operations { }; }; }; + typesafe_proxy_route_typesafe__endpoint__get: { + parameters: { + query?: never; + header?: never; + path: { + endpoint: string; + }; + cookie?: never; + }; + requestBody?: never; + responses: { + /** @description Successful Response */ + 200: { + headers: { + [name: string]: unknown; + }; + content: { + "application/json": unknown; + }; + }; + /** @description Validation Error */ + 422: { + headers: { + [name: string]: unknown; + }; + content: { + "application/json": components["schemas"]["HTTPValidationError"]; + }; + }; + }; + }; + typesafe_proxy_route_typesafe__endpoint__post: { + parameters: { + query?: never; + header?: never; + path: { + endpoint: string; + }; + cookie?: never; + }; + requestBody?: never; + responses: { + /** @description Successful Response */ + 200: { + headers: { + [name: string]: unknown; + }; + content: { + "application/json": unknown; + }; + }; + /** @description Validation Error */ + 422: { + headers: { + [name: string]: unknown; + }; + content: { + "application/json": components["schemas"]["HTTPValidationError"]; + }; + }; + }; + }; update_default_team_settings_update_default_team_settings_patch: { parameters: { query?: never; diff --git a/uv.lock b/uv.lock index 35eaa20c39e..a5e60c68515 100644 --- a/uv.lock +++ b/uv.lock @@ -10,7 +10,7 @@ resolution-markers = [ ] [options] -exclude-newer = "2026-09-14T20:32:38.482736111Z" +exclude-newer = "2026-09-14T23:55:55.024292355Z" exclude-newer-span = "P3D" [manifest] @@ -9262,11 +9262,11 @@ wheels = [ [[package]] name = "soupsieve" -version = "2.8.4" +version = "2.9.2" source = { registry = "https://pypi.org/simple" } -sdist = { url = "https://files.pythonhosted.org/packages/47/2c/0a5f6f8ee0d5589e48c7640213ed5175d52cf540a06725b628cc1a45d6ce/soupsieve-2.8.4.tar.gz", hash = "sha256:e121fd02e975c695e4e9e8774a5ee35d74714b59307868dcc5319ad2d9e3328e", size = 121110, upload-time = "2026-05-24T13:55:57.154Z" } +sdist = { url = "https://files.pythonhosted.org/packages/69/99/a6ca3beb3ccacb41fb3321d8a60e5566f9e6467601ef8eba6a17e1b89778/soupsieve-2.9.2.tar.gz", hash = "sha256:4a55d8cf158a9c2e587fa4922f1bbb91d68ac829e2d6f25403a85747c71daf74", size = 122445, upload-time = "2026-08-07T00:57:24.801Z" } wheels = [ - { url = "https://files.pythonhosted.org/packages/5e/f5/0c41cb68dcae6b7de4fac4188a3a9589e21fb31df21ea3a2e888db95e6c9/soupsieve-2.8.4-py3-none-any.whl", hash = "sha256:e7e6b0769c8f51ed59acab6e994b00621096cfb1c640a7509295987388fbaf65", size = 37304, upload-time = "2026-05-24T13:55:55.406Z" }, + { url = "https://files.pythonhosted.org/packages/eb/dc/ad025c1ee131eba60c69f4dd5779b18fcf1e6b21a343e2162a84d5d133c7/soupsieve-2.9.2-py3-none-any.whl", hash = "sha256:8089a26fd974ca7a1f30276d3d8492ab266ab15af581642dfe8aa162e0c1c823", size = 37370, upload-time = "2026-08-07T00:57:23.524Z" }, ] [[package]]