Added fix to streaming response

main/routes.py

@@ -281,10 +281,22 @@ async def generate_stream(request: GenerateRequest):
     """Generate streaming text response from prompt"""
     logger.info(f"Received streaming generation request for prompt: {request.prompt[:50]}...")
     try:
-        return api.generate_stream(
-            prompt=request.prompt,
-            system_message=request.system_message,
-            max_new_tokens=request.max_new_tokens or api.max_new_tokens
+        async def event_generator():
+            async for chunk in api.generate_stream(
+                    prompt=request.prompt,
+                    system_message=request.system_message,
+                    max_new_tokens=request.max_new_tokens or api.max_new_tokens
+            ):
+                yield f"data: {chunk}\n\n"
+            yield "data: [DONE]\n\n"
+
+        return StreamingResponse(
+            event_generator(),
+            media_type="text/event-stream",
+            headers={
+                "Cache-Control": "no-cache",
+                "Connection": "keep-alive",
+            }
         )
     except Exception as e:
         logger.error(f"Error in generate_stream endpoint: {str(e)}")