feat: Add suffix to openai_completions (#2449)

For code completion apps need "fill in the middle" capabilities. 
Added option of `suffix` to `openai_completion` to enable this. 
Updated ollama provider to showcase the same. 

### Test Plan 
```
pytest -sv --stack-config="inference=ollama"  tests/integration/inference/test_openai_completion.py --text-model qwen2.5-coder:1.5b -k test_openai_completion_non_streaming_suffix
```

### OpenAI Sample script
```
from openai import OpenAI

client = OpenAI(base_url="http://localhost:8321/v1/openai/v1")

response = client.completions.create(
    model="qwen2.5-coder:1.5b",
    prompt="The capital of ",
    suffix="is Paris.",
    max_tokens=10,
)

print(response.choices[0].text)
``` 
### Output
```
France is ____.

To answer this question, we 
```

llama_stack/apis/inference/inference.py

@@ -1038,6 +1038,8 @@ class InferenceProvider(Protocol):
         # vLLM-specific parameters
         guided_choice: list[str] | None = None,
         prompt_logprobs: int | None = None,
+        # for fill-in-the-middle type completion
+        suffix: str | None = None,
     ) -> OpenAICompletion:
         """Generate an OpenAI-compatible completion for the given prompt using the specified model.
 
@@ -1058,6 +1060,7 @@ class InferenceProvider(Protocol):
         :param temperature: (Optional) The temperature to use.
         :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.
         :returns: An OpenAICompletion.
         """
         ...

llama_stack/distribution/routers/inference.py

@@ -426,6 +426,7 @@ class InferenceRouter(Inference):
         user: str | None = None,
         guided_choice: list[str] | None = None,
         prompt_logprobs: int | None = None,
+        suffix: str | None = None,
     ) -> OpenAICompletion:
         logger.debug(
             f"InferenceRouter.openai_completion: {model=}, {stream=}, {prompt=}",
@@ -456,6 +457,7 @@ class InferenceRouter(Inference):
             user=user,
             guided_choice=guided_choice,
             prompt_logprobs=prompt_logprobs,
+            suffix=suffix,
         )
 
         provider = self.routing_table.get_provider_impl(model_obj.identifier)

llama_stack/providers/remote/inference/fireworks/fireworks.py

@@ -318,6 +318,7 @@ class FireworksInferenceAdapter(ModelRegistryHelper, Inference, NeedsRequestProv
         user: str | None = None,
         guided_choice: list[str] | None = None,
         prompt_logprobs: int | None = None,
+        suffix: str | None = None,
     ) -> OpenAICompletion:
         model_obj = await self.model_store.get_model(model)
 

llama_stack/providers/remote/inference/nvidia/nvidia.py

@@ -316,6 +316,7 @@ class NVIDIAInferenceAdapter(Inference, ModelRegistryHelper):
         user: str | None = None,
         guided_choice: list[str] | None = None,
         prompt_logprobs: int | None = None,
+        suffix: str | None = None,
     ) -> OpenAICompletion:
         provider_model_id = await self._get_provider_model_id(model)
 

llama_stack/providers/remote/inference/ollama/ollama.py

@@ -440,6 +440,7 @@ class OllamaInferenceAdapter(
         user: str | None = None,
         guided_choice: list[str] | None = None,
         prompt_logprobs: int | None = None,
+        suffix: str | None = None,
     ) -> OpenAICompletion:
         if not isinstance(prompt, str):
             raise ValueError("Ollama does not support non-string prompts for completion")
@@ -463,6 +464,7 @@ class OllamaInferenceAdapter(
             temperature=temperature,
             top_p=top_p,
             user=user,
+            suffix=suffix,
         )
         return await self.openai_client.completions.create(**params)  # type: ignore
 

llama_stack/providers/remote/inference/openai/openai.py

@@ -90,6 +90,7 @@ class OpenAIInferenceAdapter(LiteLLMOpenAIMixin):
         user: str | None = None,
         guided_choice: list[str] | None = None,
         prompt_logprobs: int | None = None,
+        suffix: str | None = None,
     ) -> OpenAICompletion:
         if guided_choice is not None:
             logging.warning("guided_choice is not supported by the OpenAI API. Ignoring.")
@@ -117,6 +118,7 @@ class OpenAIInferenceAdapter(LiteLLMOpenAIMixin):
             temperature=temperature,
             top_p=top_p,
             user=user,
+            suffix=suffix,
         )
         return await self._openai_client.completions.create(**params)
 

llama_stack/providers/remote/inference/passthrough/passthrough.py

@@ -242,6 +242,7 @@ class PassthroughInferenceAdapter(Inference):
         user: str | None = None,
         guided_choice: list[str] | None = None,
         prompt_logprobs: int | None = None,
+        suffix: str | None = None,
     ) -> OpenAICompletion:
         client = self._get_client()
         model_obj = await self.model_store.get_model(model)

llama_stack/providers/remote/inference/together/together.py

@@ -299,6 +299,7 @@ class TogetherInferenceAdapter(ModelRegistryHelper, Inference, NeedsRequestProvi
         user: str | None = None,
         guided_choice: list[str] | None = None,
         prompt_logprobs: int | None = None,
+        suffix: str | None = None,
     ) -> OpenAICompletion:
         model_obj = await self.model_store.get_model(model)
         params = await prepare_openai_completion_params(

llama_stack/providers/remote/inference/vllm/vllm.py

@@ -559,6 +559,7 @@ class VLLMInferenceAdapter(Inference, ModelsProtocolPrivate):
         user: str | None = None,
         guided_choice: list[str] | None = None,
         prompt_logprobs: int | None = None,
+        suffix: str | None = None,
     ) -> OpenAICompletion:
         self._lazy_initialize_client()
         model_obj = await self._get_model(model)

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

@@ -292,6 +292,7 @@ class WatsonXInferenceAdapter(Inference, ModelRegistryHelper):
         user: str | None = None,
         guided_choice: list[str] | None = None,
         prompt_logprobs: int | None = None,
+        suffix: str | None = None,
     ) -> OpenAICompletion:
         model_obj = await self.model_store.get_model(model)
         params = await prepare_openai_completion_params(

llama_stack/providers/utils/inference/litellm_openai_mixin.py

@@ -325,6 +325,7 @@ class LiteLLMOpenAIMixin(
         user: str | None = None,
         guided_choice: list[str] | None = None,
         prompt_logprobs: int | None = None,
+        suffix: str | None = None,
     ) -> OpenAICompletion:
         model_obj = await self.model_store.get_model(model)
         params = await prepare_openai_completion_params(

llama_stack/providers/utils/inference/openai_compat.py

@@ -1290,6 +1290,7 @@ class OpenAICompletionToLlamaStackMixin:
         user: str | None = None,
         guided_choice: list[str] | None = None,
         prompt_logprobs: int | None = None,
+        suffix: str | None = None,
     ) -> OpenAICompletion:
         if stream:
             raise ValueError(f"{self.__class__.__name__} doesn't support streaming openai completions")