From fc091c1248e3cb3276fbebeda2a8ad40aae56bf5 Mon Sep 17 00:00:00 2001
From: mateo-berri <277851410+mateo-berri@users.noreply.github.com>
Date: Tue, 1 Sep 2026 11:04:25 -0700
Subject: [PATCH] fix(prometheus): keep team alias and team wildcard names out
 of the other bucket

---
 litellm/integrations/prometheus.py            | 16 ++++++--
 ..._prometheus_requested_model_cardinality.py | 38 +++++++++++++++++++
 2 files changed, 50 insertions(+), 4 deletions(-)

diff --git a/litellm/integrations/prometheus.py b/litellm/integrations/prometheus.py
index 29b7419a91b..651f9c5d392 100644
--- a/litellm/integrations/prometheus.py
+++ b/litellm/integrations/prometheus.py
@@ -169,10 +169,11 @@ def _get_proxy_llm_router() -> Router | None:
 def _bounded_requested_model_label(requested_model: str | None) -> str | None:
     """
     Bound ``requested_model`` label cardinality: names the router recognizes
-    (model names, deployment ids, aliases, routing groups) or matches via a
-    wildcard/pattern route keep their own label value; any other
-    client-supplied string collapses into the single ``other`` bucket. With no
-    router to vouch for the string, it also collapses to ``other``.
+    (model names, deployment ids, aliases, routing groups, team public model
+    names) or matches via a global or team wildcard/pattern route keep their
+    own label value; any other client-supplied string collapses into the
+    single ``other`` bucket. With no router to vouch for the string, it also
+    collapses to ``other``.
     """
     if not requested_model:
         return requested_model
@@ -181,8 +182,15 @@ def _bounded_requested_model_label(requested_model: str | None) -> str | None:
         return UNRECOGNIZED_REQUESTED_MODEL_LABEL
     if llm_router.is_recognized_model(requested_model):
         return requested_model
+    if requested_model in llm_router.team_public_model_names:
+        return requested_model
     if llm_router.pattern_router.route(requested_model) is not None:
         return requested_model
+    if any(
+        team_pattern_router.route(requested_model) is not None
+        for team_pattern_router in llm_router.team_pattern_routers.values()
+    ):
+        return requested_model
     return UNRECOGNIZED_REQUESTED_MODEL_LABEL
 
 
diff --git a/tests/test_litellm/integrations/test_prometheus_requested_model_cardinality.py b/tests/test_litellm/integrations/test_prometheus_requested_model_cardinality.py
index 2343384e762..8d803eeab18 100644
--- a/tests/test_litellm/integrations/test_prometheus_requested_model_cardinality.py
+++ b/tests/test_litellm/integrations/test_prometheus_requested_model_cardinality.py
@@ -58,6 +58,24 @@ def router():
     )
 
 
+@pytest.fixture
+def team_router():
+    return litellm.Router(
+        model_list=[
+            {
+                "model_name": "team-internal-gpt",
+                "litellm_params": {"model": "openai/gpt-4o-mini", "api_key": "fake-key"},
+                "model_info": {"team_id": "team-1", "team_public_model_name": "team-alias-gpt"},
+            },
+            {
+                "model_name": "team-internal-bedrock",
+                "litellm_params": {"model": "openai/*", "api_key": "fake-key"},
+                "model_info": {"team_id": "team-1", "team_public_model_name": "team-models/*"},
+            },
+        ]
+    )
+
+
 def _requested_model_values(metric) -> set[str]:
     index = metric._labelnames.index("requested_model")
     return {sample_key[index] for sample_key in metric._metrics}
@@ -118,6 +136,26 @@ async def test_known_alias_and_wildcard_models_keep_their_own_labels(router):
         }
 
 
+@pytest.mark.asyncio
+async def test_team_alias_and_team_wildcard_models_keep_their_own_labels(team_router):
+    logger = PrometheusLogger()
+
+    with patch("litellm.proxy.proxy_server.llm_router", team_router, create=True):  # test-quality-ok: production reads proxy_server.llm_router lazily, no injection seam
+        await _fire_proxy_failure(logger, "team-alias-gpt")
+        await _fire_proxy_failure(logger, "team-models/gpt-4o-audio-preview")
+        await _fire_proxy_failure(logger, "agent-typo-hallucinated")
+
+    for metric in (
+        logger.litellm_proxy_failed_requests_metric,
+        logger.litellm_proxy_total_requests_metric,
+    ):
+        assert _requested_model_values(metric) == {
+            "team-alias-gpt",
+            "team-models/gpt-4o-audio-preview",
+            UNRECOGNIZED_REQUESTED_MODEL_LABEL,
+        }
+
+
 @pytest.mark.asyncio
 async def test_unknown_models_collapse_to_other_when_router_is_unavailable():
     logger = PrometheusLogger()