fix close

InternLM · Mar 7, 2024 · 6a77145 · 6a77145
1 parent d10f100
commit 6a77145
Showing 1 changed file with 3 additions and 2 deletions.
diff --git a/lmdeploy/serve/async_engine.py b/lmdeploy/serve/async_engine.py
@@ -123,10 +123,10 @@ def stream_response(self):
             self.history.append((self._prompt, resp.text))
 
     def close(self):
-        _ = self.response()
         if self._engine:
             inst = self._engine.create_instance()
-            inst.end(self._id)
+            inst.cancel(self._id)
+        _ = self.response()
 
     def __repr__(self) -> str:
         _ = self.response()
@@ -668,6 +668,7 @@ def _work_thread():
                         gen_config=gen_config,
                         stream_output=True):
                     _, res, tokens = outputs
+                    print(session._id, tokens)
                     response, state = self.tokenizer.detokenize_incrementally(
                         res,
                         state,