fix swift deploy (lmdeploy stream) (#2397)

swift/llm/deploy.py

@@ -516,9 +516,7 @@ async def _generate_full():
             prompt_tokens=num_prompt_tokens,
             completion_tokens=num_generated_tokens,
             total_tokens=num_prompt_tokens + num_generated_tokens)
-        finish_reason = None
-        if output.status.name == 'FINISH':
-            finish_reason = 'stop'
+        finish_reason = 'stop' if output.status.name == 'FINISH' else None
 
         if isinstance(request, ChatCompletionRequest):
             action, action_input = split_action_action_input(response)
@@ -554,31 +552,25 @@ async def _generate_stream():
         async with llm_engine.safe_run(session_id):
             async_iter = generator.async_stream_infer(
                 session_id=session_id, **inputs, stream_output=True, gen_config=generation_config).__aiter__()
-            is_finished = False
             response = None
-            while not is_finished:
-                try:
-                    output = await async_iter.__anext__()
-                except StopAsyncIteration:
-                    is_finished = True
+            async for output in async_iter:
                 num_generated_tokens = len(output.token_ids)
                 usage_info = UsageInfo(
                     prompt_tokens=num_prompt_tokens,
                     completion_tokens=num_generated_tokens,
                     total_tokens=num_prompt_tokens + num_generated_tokens,
                 )
+                is_finished = output.status.name == 'FINISH'
                 delta_text = template.generate_ids_to_response(
                     output.token_ids, is_finished, return_delta=True, print_idx=print_idx)
 
-                finish_reason = None
-                if output.status.name == 'FINISH':
-                    finish_reason = 'stop'
-                if not delta_text and finish_reason != 'stop':
+                if not delta_text and not is_finished:
                     continue
+                finish_reason = 'stop' if is_finished else None
                 total_response += delta_text
                 if isinstance(request, ChatCompletionRequest):
                     toolcall = None
-                    if finish_reason == 'stop':
+                    if is_finished:
                         action, action_input = split_action_action_input(total_response)
                         if action is not None:
                             toolcall = [