fix: implement incremental streaming deltas for vLLM chat completions

- Track previous_text to calculate deltas instead of sending full accumulated text
- Fixes WebUI streaming issue where responses appeared empty
- Only send new tokens in each SSE chunk delta
- Resolves OpenAI API compatibility for streaming chat completions

models/vllm/server.py

@@ -239,7 +239,11 @@ async def create_chat_completion(request: ChatCompletionRequest):
     if request.stream:
         # Streaming response
         async def generate_stream():
+            previous_text = ""
             async for output in engine.generate(prompt, sampling_params, request_id):
+                current_text = output.outputs[0].text
+                delta_text = current_text[len(previous_text):]
+                previous_text = current_text
                 chunk = {
                     "id": request_id,
                     "object": "chat.completion.chunk",
@@ -248,7 +252,7 @@ async def create_chat_completion(request: ChatCompletionRequest):
                     "choices": [
                         {
                             "index": 0,
-                            "delta": {"content": output.outputs[0].text},
+                            "delta": {"content": delta_text} if delta_text else {},
                             "finish_reason": output.outputs[0].finish_reason,
                         }
                     ]