[Bugfix] Free cross attention block table for preempted-for-recompute…

… sequence group. (vllm-project#10013) Signed-off-by: Kathy Yu <feiyangyu@google.com>
abmfy · Jan 24, 2025 · 9071af8 · 9071af8
1 parent 3712641
commit 9071af8
Showing 1 changed file with 1 addition and 0 deletions.
diff --git a/vllm/core/scheduler.py b/vllm/core/scheduler.py
@@ -1579,6 +1579,7 @@ def _preempt_by_recompute(
             seq.status = SequenceStatus.WAITING
             self.free_seq(seq)
             seq.reset_state_for_recompute()
+        self._free_seq_group_cross_attn_blocks(seq_group)
 
     def _preempt_by_swap(
         self,