xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-01-21 01:24:28 +08:00
Code Issues Packages Projects Releases Wiki Activity

[BugFix] Fix incremental detokenization perf issue (#16963)

Browse Source 
Signed-off-by: Nick Hill <nhill@redhat.com>
This commit is contained in:
Nick Hill 2025-04-22 01:16:19 -07:00 committed by GitHub
parent 8d32dc603d
commit e4d6144232
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 1 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
vllm/v1/engine/detokenizer.py
View File

@ -161,7 +161,7 @@ class FastIncrementalDetokenizer(BaseIncrementalDetokenizer):
prompt_suffix = request.prompt_token_ids
prompt_len = len(prompt_suffix)
if prompt_len > 4:
for i in range(4, max(prompt_len + 1, 32)):
for i in range(4, min(prompt_len + 1, 24)):
suffix = request.prompt_token_ids[-i:]
if '<EFBFBD>' not in self.tokenizer.decode(suffix):
prompt_suffix = suffix