featu: support passing "extra body" throught to providers

# What does this PR do?
Allows passing through extra_body parameters to inference providers.


closes #2720

## Test Plan
CI and added new test

llama_stack/apis/inference/inference.py

@@ -1058,8 +1058,6 @@ class OpenAICompletionRequest(BaseModel):
     :param top_p: (Optional) The top p to use.
     :param user: (Optional) The user to use.
     :param suffix: (Optional) The suffix that should be appended to the completion.
-    :param guided_choice: (Optional) vLLM-specific parameter for guided generation with a list of choices.
-    :param prompt_logprobs: (Optional) vLLM-specific parameter for number of log probabilities to return for prompt tokens.
     """
 
     model_config = ConfigDict(extra="allow")
@@ -1082,12 +1080,6 @@ class OpenAICompletionRequest(BaseModel):
     temperature: float | None = None
     top_p: float | None = None
     user: str | None = None
-
-    # vLLM-specific parameters (documented here but also allowed via extra fields)
-    guided_choice: list[str] | None = None
-    prompt_logprobs: int | None = None
-
-    # for fill-in-the-middle type completion
     suffix: str | None = None
 
 

llama_stack/providers/utils/inference/litellm_openai_mixin.py

@@ -230,6 +230,9 @@ class LiteLLMOpenAIMixin(
     ) -> OpenAICompletion:
         model_obj = await self.model_store.get_model(params.model)
 
+        # Extract extra fields
+        extra_body = dict(params.__pydantic_extra__ or {})
+
         request_params = await prepare_openai_completion_params(
             model=self.get_litellm_model_name(model_obj.provider_resource_id),
             prompt=params.prompt,
@@ -248,11 +251,10 @@ class LiteLLMOpenAIMixin(
             temperature=params.temperature,
             top_p=params.top_p,
             user=params.user,
-            guided_choice=params.guided_choice,
-            prompt_logprobs=params.prompt_logprobs,
             suffix=params.suffix,
             api_key=self.get_api_key(),
             api_base=self.api_base,
+            **extra_body,
         )
         return await litellm.atext_completion(**request_params)
 
@@ -272,6 +274,9 @@ class LiteLLMOpenAIMixin(
 
         model_obj = await self.model_store.get_model(params.model)
 
+        # Extract extra fields
+        extra_body = dict(params.__pydantic_extra__ or {})
+
         request_params = await prepare_openai_completion_params(
             model=self.get_litellm_model_name(model_obj.provider_resource_id),
             messages=params.messages,
@@ -298,6 +303,7 @@ class LiteLLMOpenAIMixin(
             user=params.user,
             api_key=self.get_api_key(),
             api_base=self.api_base,
+            **extra_body,
         )
         return await litellm.acompletion(**request_params)
 

llama_stack/providers/utils/inference/openai_mixin.py

@@ -228,15 +228,8 @@ class OpenAIMixin(NeedsRequestProviderData, ABC, BaseModel):
         """
         Direct OpenAI completion API call.
         """
-        # Handle parameters that are not supported by OpenAI API, but may be by the provider
-        #  prompt_logprobs is supported by vLLM
-        #  guided_choice is supported by vLLM
-        # TODO: test coverage
-        extra_body: dict[str, Any] = {}
-        if params.prompt_logprobs is not None and params.prompt_logprobs >= 0:
-            extra_body["prompt_logprobs"] = params.prompt_logprobs
-        if params.guided_choice:
-            extra_body["guided_choice"] = params.guided_choice
+        # Extract extra fields using Pydantic's built-in __pydantic_extra__
+        extra_body = dict(params.__pydantic_extra__ or {})
 
         # TODO: fix openai_completion to return type compatible with OpenAI's API response
         completion_kwargs = await prepare_openai_completion_params(
@@ -316,6 +309,9 @@ class OpenAIMixin(NeedsRequestProviderData, ABC, BaseModel):
             user=params.user,
         )
 
+        # Extract any additional provider-specific parameters using Pydantic's __pydantic_extra__
+        if extra_body := dict(params.__pydantic_extra__ or {}):
+            request_params["extra_body"] = extra_body
         resp = await self.client.chat.completions.create(**request_params)
 
         return await self._maybe_overwrite_id(resp, params.stream)  # type: ignore[no-any-return]