xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-13 21:55:32 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Hardware][Gaudi][Bugfix] Fix error for guided decoding (#12317)

Browse Source
This commit is contained in:
Yu-Zhou 2025-02-14 20:36:49 +08:00 committed by GitHub
parent 7734e9a291
commit c9e2d644e7
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 10 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

9
vllm/model_executor/guided_decoding/outlines_logits_processors.py
View File

@ -32,6 +32,8 @@ from outlines_core.fsm.json_schema import build_regex_from_schema
from pydantic import BaseModel from pydantic import BaseModel
from transformers import PreTrainedTokenizerBase from transformers import PreTrainedTokenizerBase
from vllm.platforms import current_platform
class BaseLogitsProcessor: class BaseLogitsProcessor:
@ -91,6 +93,13 @@ class BaseLogitsProcessor:
allowed_tokens = allowed_tokens.masked_select( allowed_tokens = allowed_tokens.masked_select(
allowed_tokens < scores.shape[-1]) allowed_tokens < scores.shape[-1])
mask.index_fill_(0, allowed_tokens, 0) mask.index_fill_(0, allowed_tokens, 0)
if current_platform.is_hpu():
# Workaround for HPU bug where add_() raise RuntimeError:
# synNodeCreateWithId failed for node: strided_insert
# with synStatus 1 [Invalid argument], hopefully it will
# be fixed in the future releases of the HPU runtime.
scores = scores.add(mask)
else:
scores.add_(mask) scores.add_(mask)
return scores return scores