update oai

qwen_agent/llm/qwen_oai.py

@@ -374,7 +374,8 @@ def predict(
     chunk = ChatCompletionResponse(
         model=model_id, choices=[choice_data], object="chat.completion.chunk"
     )
-    yield "{}".format(_dump_json(chunk, exclude_unset=True))
+    # yield "{}".format(_dump_json(chunk, exclude_unset=True))
+    yield chunk
 
     current_length = 0
     stop_words_ids = [tokenizer.encode(s) for s in stop_words] if stop_words else None
@@ -400,7 +401,8 @@ def predict(
         chunk = ChatCompletionResponse(
             model=model_id, choices=[choice_data], object="chat.completion.chunk"
         )
-        yield "{}".format(_dump_json(chunk, exclude_unset=True))
+        # yield "{}".format(_dump_json(chunk, exclude_unset=True))
+        yield chunk
 
     choice_data = ChatCompletionResponseStreamChoice(
         index=0, delta=DeltaMessage(), finish_reason="stop"
@@ -408,8 +410,9 @@ def predict(
     chunk = ChatCompletionResponse(
         model=model_id, choices=[choice_data], object="chat.completion.chunk"
     )
-    yield "{}".format(_dump_json(chunk, exclude_unset=True))
-    yield "[DONE]"
+    # yield "{}".format(_dump_json(chunk, exclude_unset=True))
+    yield chunk
+    # yield "[DONE]"
 
     _gc()
 
@@ -452,7 +455,6 @@ class QwenChatAsOAI(BaseChatModel):
         response = create_chat_completion(_request)
         # TODO: error handling
         for chunk in response:
-            chunk = json.loads(chunk)
             if hasattr(chunk.choices[0].delta, 'content'):
                 yield chunk.choices[0].delta.content