xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-16 09:37:25 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Bugfix][V0] Multi-sequence logprobs streaming edge case (#15259)

Browse Source 
Signed-off-by: Andy Lo <andy@mistral.ai>
This commit is contained in:
Andy Lo 2025-03-22 05:35:37 +00:00 committed by GitHub
parent 4c69e228b3
commit df1430265c
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 6 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
vllm/outputs.py
View File

@ -223,7 +223,12 @@ class RequestOutput:
if delta:
# Slice logprobs delta if applicable
if output_logprobs:
output_logprobs = output_logprobs[-num_output_tokens:]
# num_output_tokens can be 0 when n > 1 and request finishes
# before the others
if num_output_tokens > 0:
output_logprobs = output_logprobs[-num_output_tokens:]
else:
output_logprobs = None
# Don't include prompt if this is after the first output
# containing decode token ids
if include_prompt and seq.get_output_len() > num_output_tokens: