A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 15, 03:30 UTC
[Bugfix] Prevent IndexError for cached requests when pipeline parallelism is disabled (#20486)
Signed-off-by: Peter Pan <Peter.Pan@daocloud.io>
Peter Pan committed
Jul 8, 2025 at 00:41 UTC
edd270bc781e5a3d250827ac7039d71b8562bb04
1 file changed
+2
vllm/v1/core/sched/scheduler.py
+2
@@ -635,6 +635,8 @@ class Scheduler(SchedulerInterface):
635
token_ids = req.all_token_ids[req.num_computed_tokens:req.
636
num_computed_tokens + num_tokens]
637
new_token_ids.append(token_ids)
638
+ else:
639
+ new_token_ids.append([])
640
new_block_ids.append(req_to_new_block_ids[req_id])
641
num_computed_tokens.append(req.num_computed_tokens)
642
# Because resumed_reqs is usually empty, it is more efficient to do