xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-10 13:44:58 +08:00
Code Issues Packages Projects Releases Wiki Activity

vllm fix check on max vocab size (#22471)

Browse Source 
Signed-off-by: Roger Wang <hey@rogerw.io>
Signed-off-by: Roger Wang <hey@rogerw.me>
Co-authored-by: Roger Wang <hey@rogerw.io>
Co-authored-by: Roger Wang <hey@rogerw.me>
This commit is contained in:
Xiaodong Wang 2025-08-31 05:57:05 -07:00 committed by GitHub
parent 9701352e4b
commit 81eea3d348
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 13 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

14
vllm/v1/engine/processor.py
View File

@ -470,7 +470,19 @@ class Processor:
else: else:
tokenizer = self.tokenizer.get_lora_tokenizer(lora_request) tokenizer = self.tokenizer.get_lora_tokenizer(lora_request)
max_input_id = max(prompt_ids, default=0) max_input_id = max(prompt_ids, default=0)
if max_input_id > tokenizer.max_token_id:
# NOTE: tokenizer.max_token_id is the tokenizer’s vocab size while
# self.model_config.get_vocab_size() is the model’s vocab size.
# For Qwen3 models, the language model has extra tokens that do
# not exist in the tokenizer, and vice versa for multimodal
# placeholder tokens in some multimodal models.
# See https://github.com/QwenLM/Qwen3/issues/29#issuecomment-1933720399 # noqa: E501
# and https://github.com/vllm-project/vllm/pull/22471#discussion_r2312251421 # noqa: E501
# Here we take the max of the two to determine if a token id is
# truly out-of-vocabulary.
if max_input_id > max(tokenizer.max_token_id,
self.model_config.get_vocab_size() - 1):
raise ValueError( raise ValueError(
f"Token id {max_input_id} is out of vocabulary") f"Token id {max_input_id} is out of vocabulary")