xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-16 16:25:01 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Bugfix][Multi Modal] Fix incorrect Molmo token processing (#26873)

Browse Source 
Signed-off-by: sanghol <sanghol@allenai.org>
This commit is contained in:
sangho.lee 2025-10-15 02:13:59 -05:00 committed by GitHub
parent f0862eae43
commit 8865da157b
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 5 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
vllm/model_executor/models/molmo.py
View File

@ -1264,13 +1264,16 @@ class MolmoMultiModalProcessor(BaseMultiModalProcessor[MolmoProcessingInfo]):
) -> list[int]: ) -> list[int]:
processor = self.info.get_hf_processor() processor = self.info.get_hf_processor()
# Apply the chat template to the tokens # The chat template is already applied to the prompt tokens
# Use message_format="none" to avoid applying it again
# Prepend an empty space if `always_start_with_space` is True
tokens = processor.processor.get_tokens_input( # type: ignore tokens = processor.processor.get_tokens_input( # type: ignore
self.info.get_tokenizer().decode(prompt_tokens), self.info.get_tokenizer().decode(prompt_tokens),
message_format=processor.message_format, message_format="none",
always_start_with_space=processor.always_start_with_space, always_start_with_space=processor.always_start_with_space,
) )
# Prepend a BOS token id to the tokens
processed_data = self.info.ctx.call_hf_processor( processed_data = self.info.ctx.call_hf_processor(
processor, # type: ignore processor, # type: ignore
dict(tokens=tokens), dict(tokens=tokens),