Enable streaming usage metrics for OpenAI providers

Inject stream_options for telemetry, add completion streaming metrics,
fix params mutation, remove duplicate provider logic. Add unit tests.

src/llama_stack/core/routers/inference.py

@@ -176,7 +176,7 @@ class InferenceRouter(Inference):
     async def openai_completion(
         self,
         params: Annotated[OpenAICompletionRequestWithExtraBody, Body(...)],
-    ) -> OpenAICompletion:
+    ) -> OpenAICompletion | AsyncIterator[Any]:
         logger.debug(
             f"InferenceRouter.openai_completion: model={params.model}, stream={params.stream}, prompt={params.prompt}",
         )
@@ -185,9 +185,12 @@ class InferenceRouter(Inference):
         params.model = provider_resource_id
 
         if params.stream:
-            return await provider.openai_completion(params)
-            # TODO: Metrics do NOT work with openai_completion stream=True due to the fact
-            # that we do not return an AsyncIterator, our tests expect a stream of chunks we cannot intercept currently.
+            response_stream = await provider.openai_completion(params)
+            return self.wrap_completion_stream_with_metrics(
+                response=response_stream,
+                fully_qualified_model_id=request_model_id,
+                provider_id=provider.__provider_id__,
+            )
 
         response = await provider.openai_completion(params)
         response.model = request_model_id
@@ -412,16 +415,17 @@ class InferenceRouter(Inference):
                         completion_text += "".join(choice_data["content_parts"])
 
                     # Add metrics to the chunk
-                    if self.telemetry_enabled and hasattr(chunk, "usage") and chunk.usage:
-                        metrics = self._construct_metrics(
-                            prompt_tokens=chunk.usage.prompt_tokens,
-                            completion_tokens=chunk.usage.completion_tokens,
-                            total_tokens=chunk.usage.total_tokens,
-                            fully_qualified_model_id=fully_qualified_model_id,
-                            provider_id=provider_id,
-                        )
-                        for metric in metrics:
-                            enqueue_event(metric)
+                    if self.telemetry_enabled:
+                        if hasattr(chunk, "usage") and chunk.usage:
+                            metrics = self._construct_metrics(
+                                prompt_tokens=chunk.usage.prompt_tokens,
+                                completion_tokens=chunk.usage.completion_tokens,
+                                total_tokens=chunk.usage.total_tokens,
+                                fully_qualified_model_id=fully_qualified_model_id,
+                                provider_id=provider_id,
+                            )
+                            for metric in metrics:
+                                enqueue_event(metric)
 
                 yield chunk
         finally:
@@ -471,3 +475,31 @@ class InferenceRouter(Inference):
                 )
                 logger.debug(f"InferenceRouter.completion_response: {final_response}")
                 asyncio.create_task(self.store.store_chat_completion(final_response, messages))
+
+    async def wrap_completion_stream_with_metrics(
+        self,
+        response: AsyncIterator,
+        fully_qualified_model_id: str,
+        provider_id: str,
+    ) -> AsyncIterator:
+        """Stream OpenAI completion chunks and compute metrics on final chunk."""
+
+        async for chunk in response:
+            if hasattr(chunk, "model"):
+                chunk.model = fully_qualified_model_id
+
+            if getattr(chunk, "choices", None) and any(c.finish_reason for c in chunk.choices):
+                if self.telemetry_enabled:
+                    if getattr(chunk, "usage", None):
+                        usage = chunk.usage
+                        metrics = self._construct_metrics(
+                            prompt_tokens=usage.prompt_tokens,
+                            completion_tokens=usage.completion_tokens,
+                            total_tokens=usage.total_tokens,
+                            fully_qualified_model_id=fully_qualified_model_id,
+                            provider_id=provider_id,
+                        )
+                        for metric in metrics:
+                            enqueue_event(metric)
+
+            yield chunk

src/llama_stack/providers/remote/inference/bedrock/bedrock.py

@@ -8,7 +8,6 @@ from collections.abc import AsyncIterator, Iterable
 
 from openai import AuthenticationError
 
-from llama_stack.core.telemetry.tracing import get_current_span
 from llama_stack.log import get_logger
 from llama_stack.providers.utils.inference.openai_mixin import OpenAIMixin
 from llama_stack_api import (
@@ -82,14 +81,7 @@ class BedrockInferenceAdapter(OpenAIMixin):
         self,
         params: OpenAIChatCompletionRequestWithExtraBody,
     ) -> OpenAIChatCompletion | AsyncIterator[OpenAIChatCompletionChunk]:
-        """Override to enable streaming usage metrics and handle authentication errors."""
-        # Enable streaming usage metrics when telemetry is active
-        if params.stream and get_current_span() is not None:
-            if params.stream_options is None:
-                params.stream_options = {"include_usage": True}
-            elif "include_usage" not in params.stream_options:
-                params.stream_options = {**params.stream_options, "include_usage": True}
-
+        """Override to handle authentication errors and null responses."""
         try:
             logger.debug(f"Calling Bedrock OpenAI API with model={params.model}, stream={params.stream}")
             result = await super().openai_chat_completion(params=params)

src/llama_stack/providers/remote/inference/runpod/runpod.py

@@ -4,14 +4,7 @@
 # This source code is licensed under the terms described in the LICENSE file in
 # the root directory of this source tree.
 
-from collections.abc import AsyncIterator
-
 from llama_stack.providers.utils.inference.openai_mixin import OpenAIMixin
-from llama_stack_api import (
-    OpenAIChatCompletion,
-    OpenAIChatCompletionChunk,
-    OpenAIChatCompletionRequestWithExtraBody,
-)
 
 from .config import RunpodImplConfig
 
@@ -29,15 +22,3 @@ class RunpodInferenceAdapter(OpenAIMixin):
     def get_base_url(self) -> str:
         """Get base URL for OpenAI client."""
         return str(self.config.base_url)
-
-    async def openai_chat_completion(
-        self,
-        params: OpenAIChatCompletionRequestWithExtraBody,
-    ) -> OpenAIChatCompletion | AsyncIterator[OpenAIChatCompletionChunk]:
-        """Override to add RunPod-specific stream_options requirement."""
-        params = params.model_copy()
-
-        if params.stream and not params.stream_options:
-            params.stream_options = {"include_usage": True}
-
-        return await super().openai_chat_completion(params)

src/llama_stack/providers/remote/inference/watsonx/watsonx.py

@@ -10,7 +10,6 @@ from typing import Any
 import litellm
 import requests
 
-from llama_stack.core.telemetry.tracing import get_current_span
 from llama_stack.log import get_logger
 from llama_stack.providers.remote.inference.watsonx.config import WatsonXConfig
 from llama_stack.providers.utils.inference.litellm_openai_mixin import LiteLLMOpenAIMixin
@@ -56,15 +55,6 @@ class WatsonXInferenceAdapter(LiteLLMOpenAIMixin):
         Override parent method to add timeout and inject usage object when missing.
         This works around a LiteLLM defect where usage block is sometimes dropped.
         """
-
-        # Add usage tracking for streaming when telemetry is active
-        stream_options = params.stream_options
-        if params.stream and get_current_span() is not None:
-            if stream_options is None:
-                stream_options = {"include_usage": True}
-            elif "include_usage" not in stream_options:
-                stream_options = {**stream_options, "include_usage": True}
-
         model_obj = await self.model_store.get_model(params.model)
 
         request_params = await prepare_openai_completion_params(
@@ -84,7 +74,7 @@ class WatsonXInferenceAdapter(LiteLLMOpenAIMixin):
             seed=params.seed,
             stop=params.stop,
             stream=params.stream,
-            stream_options=stream_options,
+            stream_options=params.stream_options,
             temperature=params.temperature,
             tool_choice=params.tool_choice,
             tools=params.tools,

src/llama_stack/providers/utils/inference/openai_mixin.py

@@ -271,6 +271,16 @@ class OpenAIMixin(NeedsRequestProviderData, ABC, BaseModel):
         """
         Direct OpenAI completion API call.
         """
+        from llama_stack.core.telemetry.tracing import get_current_span
+
+        # inject if streaming AND telemetry active
+        if params.stream and get_current_span() is not None:
+            params = params.model_copy()
+            if params.stream_options is None:
+                params.stream_options = {"include_usage": True}
+            elif "include_usage" not in params.stream_options:
+                params.stream_options = {**params.stream_options, "include_usage": True}
+
         # TODO: fix openai_completion to return type compatible with OpenAI's API response
         provider_model_id = await self._get_provider_model_id(params.model)
         self._validate_model_allowed(provider_model_id)
@@ -308,6 +318,16 @@ class OpenAIMixin(NeedsRequestProviderData, ABC, BaseModel):
         """
         Direct OpenAI chat completion API call.
         """
+        from llama_stack.core.telemetry.tracing import get_current_span
+
+        # inject if streaming AND telemetry active
+        if params.stream and get_current_span() is not None:
+            params = params.model_copy()
+            if params.stream_options is None:
+                params.stream_options = {"include_usage": True}
+            elif "include_usage" not in params.stream_options:
+                params.stream_options = {**params.stream_options, "include_usage": True}
+
         provider_model_id = await self._get_provider_model_id(params.model)
         self._validate_model_allowed(provider_model_id)