Fix FIM for OpenRouter

FIM was not working with OpenRouter in Continue. The reason
was that we get FIM requests in `/completions`. LiteLLM when
using `acompletion` is forcing `/chat/completions` and the return
format `{..., "choices":[{"delta":{"content":"some text"}}]}`.
However, Continue was expecting the format:
`{..., "choices":[{"text":"some text"}]}` becuase of the endpoint
it called.

With this PR we force the return format to be the latter using
`atext_completion` from LiteLLM

Author: aponcedeleonch

src/codegate/providers/litellmshim/generators.py

@@ -16,20 +16,13 @@ async def sse_stream_generator(stream: AsyncIterator[Any]) -> AsyncIterator[str]
                 # alternatively we might want to just dump the whole object
                 # this might even allow us to tighten the typing of the stream
                 chunk = chunk.model_dump_json(exclude_none=True, exclude_unset=True)
-                # chunk_dict = json.loads(chunk)
-                # chunk_dict["system_fingerprint"] = "fp_00428b782a"
-                # chunk = json.dumps(chunk_dict)
             try:
-                print(f"===============\ndata: {chunk}\n\n")
                 yield f"data: {chunk}\n\n"
             except Exception as e:
-                print(f"===============\ndata: {str(e)}\n\n")
                 yield f"data: {str(e)}\n\n"
     except Exception as e:
-        print(f"===============\ndata: {str(e)}\n\n")
         yield f"data: {str(e)}\n\n"
     finally:
-        print(f"===============\ndata: [DONE]\n\n")
         yield "data: [DONE]\n\n"
 
 

src/codegate/providers/openai/provider.py

@@ -18,8 +18,9 @@ class OpenAIProvider(BaseProvider):
     def __init__(
         self,
         pipeline_factory: PipelineFactory,
+        # Enable receiving other completion handlers from childs, i.e. OpenRouter and LM Studio
+        completion_handler: LiteLLmShim = LiteLLmShim(stream_generator=sse_stream_generator),
     ):
-        completion_handler = LiteLLmShim(stream_generator=sse_stream_generator)
         super().__init__(
             OpenAIInputNormalizer(),
             OpenAIOutputNormalizer(),

src/codegate/providers/openrouter/provider.py

@@ -2,12 +2,14 @@
 from typing import Dict
 
 from fastapi import Header, HTTPException, Request
+from litellm import atext_completion
 from litellm.types.llms.openai import ChatCompletionRequest
 
 from codegate.clients.clients import ClientType
 from codegate.clients.detector import DetectClient
 from codegate.pipeline.factory import PipelineFactory
 from codegate.providers.fim_analyzer import FIMAnalyzer
+from codegate.providers.litellmshim import LiteLLmShim, sse_stream_generator
 from codegate.providers.normalizer.completion import CompletionNormalizer
 from codegate.providers.openai import OpenAIProvider
 
@@ -20,15 +22,20 @@ def normalize(self, data: Dict) -> ChatCompletionRequest:
         return super().normalize(data)
 
     def denormalize(self, data: ChatCompletionRequest) -> Dict:
-        if data.get("had_prompt_before", False):
-            del data["had_prompt_before"]
-
-        return data
+        return super().denormalize(data)
 
 
 class OpenRouterProvider(OpenAIProvider):
     def __init__(self, pipeline_factory: PipelineFactory):
-        super().__init__(pipeline_factory)
+        super().__init__(
+            pipeline_factory,
+            # We get FIM requests in /completions. LiteLLM is forcing /chat/completions
+            # which returns "choices":[{"delta":{"content":"some text"}}]
+            # instead of "choices":[{"text":"some text"}] expected by the client (Continue)
+            completion_handler=LiteLLmShim(
+                stream_generator=sse_stream_generator, fim_completion_func=atext_completion
+            ),
+        )
         self._fim_normalizer = OpenRouterNormalizer()
 
     @property