biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Bugfix] Prevent IndexError for cached requests when pipeline parallelism is disabled (#20486)

Browse Source 
Signed-off-by: Peter Pan <Peter.Pan@daocloud.io>
This commit is contained in:
Peter Pan
2025-07-08 00:41:15 +08:00
committed by GitHub
parent 110df74332
commit edd270bc78
1 changed files with 2 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
vllm/v1/core/sched/scheduler.py
View File

@@ -635,6 +635,8 @@ class Scheduler(SchedulerInterface):
token_ids = req.all_token_ids[req.num_computed_tokens:req. token_ids = req.all_token_ids[req.num_computed_tokens:req.
num_computed_tokens + num_tokens] num_computed_tokens + num_tokens]
new_token_ids.append(token_ids) new_token_ids.append(token_ids)
else:
new_token_ids.append([])
new_block_ids.append(req_to_new_block_ids[req_id]) new_block_ids.append(req_to_new_block_ids[req_id])
num_computed_tokens.append(req.num_computed_tokens) num_computed_tokens.append(req.num_computed_tokens)
# Because resumed_reqs is usually empty, it is more efficient to do # Because resumed_reqs is usually empty, it is more efficient to do