Merge branch 'lizhg1-v0.8.5.post1-dev-patch-30255' into 'v0.8.5.post1-dev'

Update sequence.py fix assert error。

See merge request dcutoolkit/deeplearing/vllm!121

vllm/zero_overhead/sequence.py

 
-
 from typing import Union
 from vllm.sequence import Sequence
 from typing import Sequence as GenericSequence
@@ -12,7 +11,7 @@ class ZeroOverheadSequence(Sequence):
 
     def fix_last_token_id(self, token_id: int) -> None:
         effect_offset = self.effective_output_len - len(self.data.output_token_ids)
-        assert effect_offset < 0
+        if effect_offset < 0:
             self.data._output_token_ids[effect_offset] = token_id
             if len(self.data._new_appended_tokens) >= effect_offset * -1:
                 self.data._new_appended_tokens[effect_offset] = token_id