minor update

Author: lvhan028

lmdeploy/pytorch/messages.py

@@ -657,7 +657,7 @@ class SchedulerSequence:
     # mrope
     history_mrope_pos_ids: HistoryMropePosIds = field(default_factory=HistoryMropePosIds)
 
-    # prefix caching
+    # Prompt tokens reused from prefix cache; set by BlockTrie.match().
     cached_tokens: int = 0
 
     def __post_init__(self):

lmdeploy/pytorch/paging/block_trie.py

@@ -83,7 +83,6 @@ def get_root(self, adapter_name: str):
     def match(self, seq: SchedulerSequence):
         """Match sequence and cache."""
         if not self.enable:
-            seq.cached_tokens = 0
             return
 
         block_size = self.block_size