fix: do not assume streaming for ollama

Aider is sending a mix of requests that need stream or sync processing
and we were hardcoding stream to True. Instead, pick if from the request
and also fix the formatting of the json response

Closes: #624

Author: yrobla

src/codegate/providers/ollama/adapter.py

@@ -25,10 +25,8 @@ def normalize(self, data: Dict) -> ChatCompletionRequest:
                 {"content": normalized_data.pop("prompt"), "role": "user"}
             ]
 
-        # In Ollama force the stream to be True. Continue is not setting this parameter and
-        # most of our functionality is for streaming completions.
-        normalized_data["stream"] = True
-
+        # if we have the stream flag in data we set it, otherwise defaults to true
+        normalized_data["stream"] = data.get("stream", True)
         return ChatCompletionRequest(**normalized_data)
 
     def denormalize(self, data: ChatCompletionRequest) -> Dict:

src/codegate/providers/ollama/completion_handler.py

@@ -60,7 +60,7 @@ def _create_streaming_response(self, stream: AsyncIterator[ChatResponse]) -> Str
         """
         return StreamingResponse(
             ollama_stream_generator(stream),
-            media_type="application/x-ndjson",
+            media_type="application/x-ndjson; charset=utf-8",
             headers={
                 "Cache-Control": "no-cache",
                 "Connection": "keep-alive",
@@ -70,4 +70,4 @@ def _create_streaming_response(self, stream: AsyncIterator[ChatResponse]) -> Str
     def _create_json_response(
         self, response: Union[GenerateResponse, ChatResponse]
     ) -> JSONResponse:
-        return JSONResponse(content=response.model_dump_json(), status_code=200)
+        return JSONResponse(status_code=200, content=response.model_dump())

src/codegate/providers/ollama/provider.py

@@ -65,7 +65,7 @@ async def show_model(request: Request):
                 response = await client.post(
                     f"{self.base_url}/api/show",
                     content=body,
-                    headers={"Content-Type": "application/json"},
+                    headers={"Content-Type": "application/json; charset=utf-8"},
                 )
                 return response.json()
 
@@ -99,4 +99,5 @@ async def create_completion(request: Request):
                 else:
                     # just continue raising the exception
                     raise e
+            
             return self._completion_handler.create_response(stream)