xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-03-31 02:07:02 +08:00
Code Issues Packages Projects Releases Wiki Activity

fix corner case for update_async_output_token_ids

Browse Source 
Signed-off-by: zhuhaoran <zhuhaoran.zhr@alibaba-inc.com>
This commit is contained in:
zhuhaoran 2025-12-23 17:52:38 +08:00
parent 699800a28e
commit 8d339e86e5
1 changed files with 6 additions and 9 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

15
vllm/v1/worker/gpu_input_batch.py
View File

@ -942,18 +942,15 @@ class InputBatch:
sampled_token_ids = self.sampled_token_ids_cpu.tolist()
# Replace placeholder token id(s) with actual sampled id(s).
if sampled_ids := sampled_token_ids[prev_index]:
num_placeholders = 0
for t in reversed(req_output_token_ids):
num_replace = 0
for t in sampled_ids:
if t == -1:
num_placeholders += 1
else:
break
if num_placeholders == 0:
num_replace += 1
if num_replace == 0:
continue
assert num_placeholders <= len(sampled_ids)
req_output_token_ids[-num_placeholders:] = sampled_ids[
:num_placeholders
]
req_output_token_ids[-num_replace:] = sampled_ids[:num_replace]
def update_async_spec_token_ids(
self,