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

[Bugfix] using len(tokenizer) instead of tokenizer.vocab_size in AllowedTokenIdsLogitsProcessor (#11156)

Browse Source
This commit is contained in:
zhangjf
2024-12-13 23:56:19 +08:00
committed by GitHub
parent c31d4a57a6
commit 5b0ed8391d
1 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
vllm/entrypoints/openai/logits_processors.py
View File

@@ -71,7 +71,7 @@ def get_logits_processors(
# Check if token_id is within the vocab size # Check if token_id is within the vocab size
for token_id, bias in clamped_logit_bias.items(): for token_id, bias in clamped_logit_bias.items():
if token_id < 0 or token_id >= tokenizer.vocab_size: if token_id < 0 or token_id >= len(tokenizer):
raise ValueError(f"token_id {token_id} in logit_bias contains " raise ValueError(f"token_id {token_id} in logit_bias contains "
"out-of-vocab token id") "out-of-vocab token id")
@@ -81,6 +81,6 @@ def get_logits_processors(
if allowed_token_ids is not None: if allowed_token_ids is not None:
logits_processors.append( logits_processors.append(
_get_allowed_token_ids_logits_processor( _get_allowed_token_ids_logits_processor(
frozenset(allowed_token_ids), tokenizer.vocab_size)) frozenset(allowed_token_ids), len(tokenizer)))
return logits_processors return logits_processors