LCORE-4704: Azure inference provider

src/providers/inference/azure/README.md

@@ -4,3 +4,6 @@
 
 Azure inference provider package.
 
+## [provider.py](provider.py)
+
+Azure OpenAI runtime inference provider backed by Pydantic AI.

src/providers/inference/azure/__init__.py

@@ -1 +1,5 @@
 """Azure inference provider package."""
+
+from providers.inference.azure.provider import AzureInferenceProvider
+
+__all__ = ["AzureInferenceProvider"]

src/providers/inference/azure/provider.py

@@ -0,0 +1,142 @@
+"""Azure OpenAI runtime inference provider backed by Pydantic AI."""
+
+from __future__ import annotations
+
+from typing import Any
+
+from azure.identity.aio import ClientSecretCredential, get_bearer_token_provider
+from openai import AsyncAzureOpenAI
+from pydantic_ai.embeddings import EmbeddingModel
+from pydantic_ai.embeddings.openai import OpenAIEmbeddingModel
+from pydantic_ai.models import Model
+from pydantic_ai.models.openai import OpenAIResponsesModel
+from pydantic_ai.providers.azure import AzureProvider
+from sentence_transformers import CrossEncoder
+
+from configuration import configuration
+from models.config import AzureInferenceProviderEntry
+from providers.inference.base import InferenceProvider
+from providers.inference.catalog import CatalogModel, DiscoveredModel
+from providers.inference.errors import ModelTypeNotSupportedError
+from providers.inference.model_type_inference import (
+    infer_model_type_from_pai_known_names,
+)
+from providers.inference.types import ModelType
+
+
+class AzureInferenceProvider(
+    InferenceProvider[AzureInferenceProviderEntry, AzureProvider]
+):
+    """Inference provider for Azure OpenAI HTTP APIs."""
+
+    def _build_provider(self) -> AzureProvider:
+        """Build the Pydantic AI Azure provider from an LCORE-owned client.
+
+        Returns:
+            A Pydantic AI Azure provider wrapping an ``AsyncAzureOpenAI``
+            client configured with API-key or Entra ID authentication.
+
+        """
+        config = self._entry.config
+        client_kwargs: dict[str, Any] = {
+            "api_version": config.api_version,
+        }
+        if config.base_url is not None:
+            client_kwargs["azure_endpoint"] = config.base_url
+
+        if (entra_id := config.entra_id or configuration.azure_entra_id) is not None:
+            credential = ClientSecretCredential(
+                tenant_id=entra_id.tenant_id.get_secret_value(),
+                client_id=entra_id.client_id.get_secret_value(),
+                client_secret=entra_id.client_secret.get_secret_value(),
+            )
+            client_kwargs["azure_ad_token_provider"] = get_bearer_token_provider(
+                credential,
+                entra_id.scope,
+            )
+        else:
+            client_kwargs["api_key"] = (
+                config.api_key.get_secret_value()
+                if config.api_key is not None
+                else None
+            )
+
+        return AzureProvider(openai_client=AsyncAzureOpenAI(**client_kwargs))
+
+    async def _discover_models(self) -> list[DiscoveredModel]:
+        """List models from the Azure OpenAI models endpoint.
+
+        Returns:
+            Discovered models with inferred types, before allow-list filtering.
+
+        """
+        response = await self._get_provider.client.models.list()
+        return [
+            DiscoveredModel(
+                provider_model_id=model.id,
+                model_type=self._infer_model_type(model.id),
+            )
+            for model in response.data
+        ]
+
+    def _create_llm(self, model: CatalogModel) -> Model:
+        """Create a Responses API chat model for a catalog entry.
+
+        Args:
+            model: Catalog entry describing the LLM to instantiate.
+
+        Returns:
+            A Pydantic AI OpenAI Responses model bound to this provider.
+
+        """
+        return OpenAIResponsesModel(
+            model.provider_model_id,
+            provider=self._get_provider,
+        )
+
+    def _create_embedding(self, model: CatalogModel) -> EmbeddingModel:
+        """Create an embedding model for a catalog entry.
+
+        Args:
+            model: Catalog entry describing the embedding model to instantiate.
+
+        Returns:
+            A Pydantic AI OpenAI embedding model bound to this provider.
+
+        """
+        return OpenAIEmbeddingModel(
+            model.provider_model_id,
+            provider=self._get_provider,
+        )
+
+    def _create_rerank(self, model: CatalogModel) -> CrossEncoder:
+        """Reject rerank requests; this provider has no rerank API.
+
+        Args:
+            model: Catalog entry that requested rerank support.
+
+        Raises:
+            ModelTypeNotSupportedError: Always, because rerank is unsupported.
+
+        """
+        _ = model
+        raise ModelTypeNotSupportedError("rerank", self.provider_id)
+
+    def _infer_model_type(
+        self,
+        provider_model_id: str,
+        **kwargs: Any,
+    ) -> ModelType:
+        """Infer model type from Pydantic AI known LLM and embedding names.
+
+        Args:
+            provider_model_id: Model id from Azure OpenAI or from an allow-list
+                entry.
+            **kwargs: Unused; accepted for base-class compatibility.
+
+        Returns:
+            embedding when the id is a known embedding model, otherwise llm.
+
+        """
+        _ = kwargs
+        return infer_model_type_from_pai_known_names(provider_model_id) or "llm"

src/providers/inference/registry.py

@@ -5,6 +5,7 @@
 from typing import Any, Final, Optional
 
 from models.config import InferenceConfiguration
+from providers.inference.azure.provider import AzureInferenceProvider
 from providers.inference.base import InferenceProvider
 from providers.inference.catalog import CatalogModel
 from providers.inference.errors import ProviderNotFoundError
@@ -14,6 +15,7 @@
 
 PROVIDER_CLASSES: Final[dict[ProviderType, type[InferenceProvider[Any, Any]]]] = {
     "openai": OpenAIInferenceProvider,
+    "azure": AzureInferenceProvider,
 }
 
 

tests/unit/providers/inference/azure/__init__.py

@@ -0,0 +1 @@
+"""Unit tests for the Azure inference provider."""