diff --git a/litellm/llms/azure/azure.py b/litellm/llms/azure/azure.py index 980b27cda55..2bcc830851a 100644 --- a/litellm/llms/azure/azure.py +++ b/litellm/llms/azure/azure.py @@ -846,6 +846,7 @@ class AzureChatCompletion(BaseAzureLLM, BaseLLM): api_key: str, data: dict, headers: dict, + deployment_name: str | None = None, ) -> httpx.Response: """ Implemented for azure dall-e-2 image gen calls @@ -957,7 +958,7 @@ class AzureChatCompletion(BaseAzureLLM, BaseLLM): content=json.dumps(result).encode("utf-8"), request=httpx.Request(method="POST", url="https://api.openai.com/v1"), ) - request_json: Final = azure_deployment_image_generation_json_body(api_base, data) + request_json: Final = azure_deployment_image_generation_json_body(api_base, data, deployment_name) return await async_handler.post( url=api_base, json=request_json, @@ -973,6 +974,7 @@ class AzureChatCompletion(BaseAzureLLM, BaseLLM): api_key: str, data: dict, headers: dict, + deployment_name: str | None = None, ) -> httpx.Response: """ Implemented for azure dall-e-2 image gen calls @@ -1073,7 +1075,7 @@ class AzureChatCompletion(BaseAzureLLM, BaseLLM): content=json.dumps(result).encode("utf-8"), request=httpx.Request(method="POST", url="https://api.openai.com/v1"), ) - request_json: Final = azure_deployment_image_generation_json_body(api_base, data) + request_json: Final = azure_deployment_image_generation_json_body(api_base, data, deployment_name) return sync_handler.post( url=api_base, json=request_json, @@ -1091,9 +1093,10 @@ class AzureChatCompletion(BaseAzureLLM, BaseLLM): AzureFoundryMAIImageGenerationConfig, ) - api_base: str = azure_client_params.get("azure_endpoint", "") # "https://example-endpoint.openai.azure.com" - if api_base.endswith("/"): - api_base = api_base.rstrip("/") + # deployment-scoped endpoints are moved to "base_url" by select_azure_base_url_or_endpoint + api_base: str = (azure_client_params.get("azure_endpoint") or azure_client_params.get("base_url") or "").rstrip( + "/" + ) api_version: Final[str] = azure_client_params.get("api_version", "") if model is None: model = "" @@ -1113,6 +1116,14 @@ class AzureChatCompletion(BaseAzureLLM, BaseLLM): api_version=api_version, ) + v1_url: Final = BaseAzureLLM.get_azure_v1_image_url( + api_base=api_base, + api_version=api_version, + route="/openai/images/generations", + ) + if v1_url is not None: + return v1_url + if "/openai/deployments/" in api_base: base_url_with_deployment = api_base else: @@ -1167,6 +1178,7 @@ class AzureChatCompletion(BaseAzureLLM, BaseLLM): api_key=api_key, data=data, headers=headers, + deployment_name=model, ) provider_config: Final = get_azure_image_generation_config(data.get("model", "dall-e-2")) @@ -1302,6 +1314,7 @@ class AzureChatCompletion(BaseAzureLLM, BaseLLM): api_key=api_key or "", data=data, headers=headers, + deployment_name=model, ) provider_config: Final = get_azure_image_generation_config(data.get("model", "dall-e-2")) if isinstance(provider_config, AzureFoundryMAIImageGenerationConfig): diff --git a/litellm/llms/azure/common_utils.py b/litellm/llms/azure/common_utils.py index 2c34851d275..6cb7d09cec4 100644 --- a/litellm/llms/azure/common_utils.py +++ b/litellm/llms/azure/common_utils.py @@ -4,6 +4,7 @@ import json import os from collections.abc import Callable, Mapping from functools import lru_cache +from types import MappingProxyType from typing import Any, Final, Literal, NamedTuple, cast import httpx @@ -789,6 +790,32 @@ class BaseAzureLLM(BaseOpenAILLM): return str(final_url) + @staticmethod + def get_azure_v1_image_url(api_base: str, api_version: str | None, route: str) -> str | None: + """ + Azure's v1 surface serves images at ``/openai/v1/images/{generations,edits}`` and routes by + ``model`` in the request body, so any deployment path and stale ``api-version`` in + ``api_base`` have to be dropped. + + Returns None when ``api_version`` is a dated one, which still uses the deployment route. + """ + if not BaseAzureLLM._is_azure_v1_api_version(api_version): + return None + + base_url: Final = httpx.URL(api_base) + openai_path_start: Final = base_url.path.find("/openai") + resource_base: Final = str( + base_url.copy_with( + path=base_url.path if openai_path_start == -1 else base_url.path[:openai_path_start], + params=httpx.QueryParams(tuple((k, v) for k, v in base_url.params.multi_items() if k != "api-version")), + ) + ) + return BaseAzureLLM._get_base_azure_url( + api_base=resource_base, + litellm_params=MappingProxyType({"api_version": api_version}), + route=route, + ) + @staticmethod def _is_azure_v1_api_version(api_version: str | None) -> bool: if api_version is None: diff --git a/litellm/llms/azure/image_edit/transformation.py b/litellm/llms/azure/image_edit/transformation.py index 15592968bad..e4716289a34 100644 --- a/litellm/llms/azure/image_edit/transformation.py +++ b/litellm/llms/azure/image_edit/transformation.py @@ -93,8 +93,6 @@ class AzureImageEditConfig(OpenAIImageEditConfig): raise ValueError( f"api_base is required for Azure AI Studio. Please set the api_base parameter. Passed `api_base={api_base}`" ) - original_url: Final = httpx.URL(api_base) - # Resolve api_version: litellm_params > litellm.api_version > AZURE_API_VERSION env > default. # Mirrors the fallback chain used by the Azure chat path in common_utils.py, # so callers that set a global / env api_version don't get an unversioned URL. @@ -105,6 +103,16 @@ class AzureImageEditConfig(OpenAIImageEditConfig): or litellm.AZURE_DEFAULT_API_VERSION ) + v1_url: Final = BaseAzureLLM.get_azure_v1_image_url( + api_base=api_base, + api_version=api_version, + route="/openai/images/edits", + ) + if v1_url is not None: + return v1_url + + original_url: Final = httpx.URL(api_base) + # Create a new dictionary with existing params query_params: Final = dict(original_url.params) diff --git a/litellm/llms/azure/image_generation/http_utils.py b/litellm/llms/azure/image_generation/http_utils.py index 03c425eeffc..1aa5757ca95 100644 --- a/litellm/llms/azure/image_generation/http_utils.py +++ b/litellm/llms/azure/image_generation/http_utils.py @@ -1,7 +1,9 @@ """HTTP helpers for Azure OpenAI image generation (REST, not SDK).""" +from typing import Final -def azure_deployment_image_generation_json_body(api_base: str, data: dict) -> dict: + +def azure_deployment_image_generation_json_body(api_base: str, data: dict, deployment_name: str | None = None) -> dict: """ Build the JSON body for Azure OpenAI image generation POSTs. @@ -9,9 +11,20 @@ def azure_deployment_image_generation_json_body(api_base: str, data: dict) -> di deployment in the URL only; sending ``model`` in the body (especially the deployment name) breaks some models (e.g. gpt-image-2). See LiteLLM #26316. + For the v1 surface (``.../openai/v1/images/...``), Azure routes by the deployment + name in the body ``model`` field, so the deployment name must replace any base + model name there or Azure answers 404 DeploymentNotFound. + Provider-style URLs (e.g. ``/providers/...`` for FLUX on Azure AI) keep all keys so non–OpenAI-deployment payloads still work. """ - if "images/generations" in api_base and "/openai/deployments/" in api_base: - return {k: v for k, v in data.items() if k != "model"} - return data + drop_model: Final = "images/generations" in api_base and "/openai/deployments/" in api_base + v1_route: Final = "/openai/v1/images/" in api_base and bool(deployment_name) + if not drop_model and not v1_route: + return data + entries: Final = ( + tuple((k, v) for k, v in data.items() if k != "model") + if drop_model + else (*data.items(), ("model", deployment_name)) + ) + return {k: v for k, v in entries} diff --git a/tests/test_litellm/llms/azure/image_edit/test_azure_image_edit_transformation.py b/tests/test_litellm/llms/azure/image_edit/test_azure_image_edit_transformation.py index 59472d1a49d..5c7b249ae72 100644 --- a/tests/test_litellm/llms/azure/image_edit/test_azure_image_edit_transformation.py +++ b/tests/test_litellm/llms/azure/image_edit/test_azure_image_edit_transformation.py @@ -233,3 +233,62 @@ def test_api_version_in_api_base_query_is_preserved(monkeypatch): ) assert _query_params(url) == {"api-version": "2024-05-01-preview"} + + +def test_v1_api_version_uses_v1_route_and_keeps_model(monkeypatch): + monkeypatch.setattr(litellm, "api_version", None, raising=False) + monkeypatch.delenv("AZURE_API_VERSION", raising=False) + config = AzureImageEditConfig() + + for api_version in ("v1", "preview", "latest"): + url = config.get_complete_url( + model=_FALLBACK_MODEL, + api_base=_FALLBACK_API_BASE, + litellm_params={"api_version": api_version}, + ) + assert urllib.parse.urlparse(url).path == "/openai/v1/images/edits" + assert _query_params(url) == {"api-version": api_version} + assert config.finalize_image_edit_request_data({"model": _FALLBACK_MODEL, "prompt": "x"}, url) == { + "model": _FALLBACK_MODEL, + "prompt": "x", + } + + +def test_v1_api_version_from_global_uses_v1_route(monkeypatch): + monkeypatch.setattr(litellm, "api_version", "preview", raising=False) + monkeypatch.delenv("AZURE_API_VERSION", raising=False) + + url = AzureImageEditConfig().get_complete_url( + model=_FALLBACK_MODEL, + api_base=_FALLBACK_API_BASE, + litellm_params={}, + ) + + assert urllib.parse.urlparse(url).path == "/openai/v1/images/edits" + + +def test_dated_api_version_still_uses_deployment_route(monkeypatch): + monkeypatch.setattr(litellm, "api_version", None, raising=False) + monkeypatch.delenv("AZURE_API_VERSION", raising=False) + + url = AzureImageEditConfig().get_complete_url( + model=_FALLBACK_MODEL, + api_base=_FALLBACK_API_BASE, + litellm_params={"api_version": "2024-10-21"}, + ) + + assert urllib.parse.urlparse(url).path == f"/openai/deployments/{_FALLBACK_MODEL}/images/edits" + + +def test_v1_api_version_replaces_deployment_scoped_api_base(monkeypatch): + monkeypatch.setattr(litellm, "api_version", None, raising=False) + monkeypatch.delenv("AZURE_API_VERSION", raising=False) + + url = AzureImageEditConfig().get_complete_url( + model=_FALLBACK_MODEL, + api_base=f"{_FALLBACK_API_BASE}/openai/deployments/{_FALLBACK_MODEL}/images/edits?api-version=2024-10-21", + litellm_params={"api_version": "preview"}, + ) + + assert urllib.parse.urlparse(url).path == "/openai/v1/images/edits" + assert _query_params(url) == {"api-version": "preview"} diff --git a/tests/test_litellm/llms/azure/image_generation/test_azure_image_generation_init.py b/tests/test_litellm/llms/azure/image_generation/test_azure_image_generation_init.py index 560fee17328..70b5eab5c37 100644 --- a/tests/test_litellm/llms/azure/image_generation/test_azure_image_generation_init.py +++ b/tests/test_litellm/llms/azure/image_generation/test_azure_image_generation_init.py @@ -3,9 +3,12 @@ import traceback from typing import Callable, Optional from unittest.mock import AsyncMock, MagicMock, Mock, patch +import httpx import pytest +import respx import litellm +from litellm.caching.llm_caching_handler import LLMClientCache from litellm.llms.azure.azure import AzureChatCompletion from litellm.llms.azure.image_generation.http_utils import ( azure_deployment_image_generation_json_body, @@ -433,3 +436,154 @@ async def test_azure_aimage_generation_base_model_vs_deployment_name(): wire_json = post_kwargs.get("json") or {} assert "model" not in wire_json assert data.get("model") == base_model + + +@pytest.mark.parametrize("api_version", ["v1", "preview", "latest"]) +def test_azure_image_generation_v1_api_version_uses_v1_route(api_version): + """The v1 Azure surface exposes /openai/v1/images/generations and routes by body ``model``.""" + url = AzureChatCompletion().create_azure_base_url( + azure_client_params={ + "azure_endpoint": "https://my-resource.openai.azure.com", + "api_version": api_version, + }, + model="gpt-image-1", + base_model=None, + ) + assert url == f"https://my-resource.openai.azure.com/openai/v1/images/generations?api-version={api_version}" + data = {"model": "gpt-image-1", "prompt": "x"} + assert azure_deployment_image_generation_json_body(url, data) == data + + +def test_azure_image_generation_dated_api_version_uses_deployment_route(): + url = AzureChatCompletion().create_azure_base_url( + azure_client_params={ + "azure_endpoint": "https://my-resource.openai.azure.com", + "api_version": "2024-10-21", + }, + model="gpt-image-1", + base_model=None, + ) + assert ( + url + == "https://my-resource.openai.azure.com/openai/deployments/gpt-image-1/images/generations?api-version=2024-10-21" + ) + assert "model" not in azure_deployment_image_generation_json_body(url, {"model": "gpt-image-1", "prompt": "x"}) + + +def test_azure_image_generation_v1_api_version_replaces_deployment_scoped_api_base(): + url = AzureChatCompletion().create_azure_base_url( + azure_client_params={ + "azure_endpoint": "https://my-resource.openai.azure.com/openai/deployments/gpt-image-1/images/generations", + "api_version": "preview", + }, + model="gpt-image-1", + base_model=None, + ) + assert url == "https://my-resource.openai.azure.com/openai/v1/images/generations?api-version=preview" + + +def test_azure_image_generation_v1_api_version_uses_base_url_client_param(): + url = AzureChatCompletion().create_azure_base_url( + azure_client_params={ + "base_url": "https://my-resource.openai.azure.com/openai/deployments/gpt-image-1?api-version=2024-10-21", + "api_version": "preview", + }, + model="gpt-image-1", + base_model=None, + ) + assert url == "https://my-resource.openai.azure.com/openai/v1/images/generations?api-version=preview" + + +def test_azure_v1_image_generation_json_body_sends_deployment_name(): + """The v1 route ignores the URL and routes by body ``model``, which must be the deployment name.""" + url = "https://my-resource.openai.azure.com/openai/v1/images/generations?api-version=preview" + data = {"model": "gpt-image-2", "prompt": "x", "n": 1} + out = azure_deployment_image_generation_json_body(url, data, deployment_name="img-dep") + assert out["model"] == "img-dep" + assert out["prompt"] == "x" + assert data["model"] == "gpt-image-2" + assert azure_deployment_image_generation_json_body(url, data) == data + + +@pytest.mark.asyncio +async def test_azure_aimage_generation_v1_route_sends_deployment_name_in_body( + respx_mock: respx.MockRouter, monkeypatch: pytest.MonkeyPatch +): + monkeypatch.setattr(litellm, "disable_aiohttp_transport", True) + monkeypatch.setattr(litellm, "in_memory_llm_clients_cache", LLMClientCache()) + azure_chat_completion = AzureChatCompletion() + model = "img-dep" + base_model = "gpt-image-2" + data = {"model": base_model, "prompt": "A beautiful image of a cat", "n": 1} + azure_client_params = { + "azure_endpoint": "https://my-resource.openai.azure.com", + "api_version": "preview", + } + + route = respx_mock.post("https://my-resource.openai.azure.com/openai/v1/images/generations").mock( + return_value=httpx.Response(200, json={"created": 1234567890, "data": [{"b64_json": "aaaa"}]}) + ) + + logging_obj = MagicMock() + logging_obj.pre_call = MagicMock() + logging_obj.post_call = MagicMock() + + await azure_chat_completion.aimage_generation( + data=data, + model_response=None, + azure_client_params=azure_client_params, + api_key="test-api-key", + input=[], + logging_obj=logging_obj, + headers={}, + model=model, + timeout=60.0, + ) + + request = route.calls.last.request + assert str(request.url) == ("https://my-resource.openai.azure.com/openai/v1/images/generations?api-version=preview") + sent_body = json.loads(request.content) + assert sent_body["model"] == model + assert sent_body["prompt"] == data["prompt"] + + +def test_azure_image_generation_v1_route_base_model_vs_deployment_name(respx_mock: respx.MockRouter): + """On the v1 surface the body ``model`` must be the deployment name, never base_model.""" + azure_chat_completion = AzureChatCompletion() + prompt = "A beautiful image of a cat" + model = "img-dep" + base_model = "gpt-image-2" + api_base = "https://my-resource.openai.azure.com" + api_version = "v1" + litellm_params = { + "base_model": base_model, + "api_base": api_base, + "api_version": api_version, + } + + route = respx_mock.post(f"{api_base}/openai/v1/images/generations").mock( + return_value=httpx.Response(200, json={"created": 1234567890, "data": [{"b64_json": "aaaa"}]}) + ) + + logging_obj = MagicMock() + logging_obj.pre_call = MagicMock() + logging_obj.post_call = MagicMock() + + azure_chat_completion.image_generation( + prompt=prompt, + timeout=60.0, + optional_params={"n": 1, "size": "1024x1024"}, + logging_obj=logging_obj, + headers={}, + model=model, + api_key="test-api-key", + api_base=api_base, + api_version=api_version, + litellm_params=litellm_params, + ) + + request = route.calls.last.request + assert str(request.url) == f"{api_base}/openai/v1/images/generations?api-version={api_version}" + sent_body = json.loads(request.content) + assert sent_body["model"] == model + assert sent_body["prompt"] == prompt