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

[Multimodal] Generate mm_hash based on request metadata when caching is turned off (#23690)

Browse Source 
Signed-off-by: Roger Wang <hey@rogerw.io>
This commit is contained in:
Roger Wang
2025-08-27 13:24:31 -07:00
committed by GitHub
parent 0585a9e73c
commit 8bf6266a17
12 changed files with 179 additions and 24 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
vllm/model_executor/models/pixtral.py
View File

@@ -314,12 +314,14 @@ class PixtralMultiModalProcessor(BaseMultiModalProcessor[PixtralProcessingInfo]
mm_data_items: MultiModalDataItems,
hf_processor_mm_kwargs: Mapping[str, object],
tokenization_kwargs: Mapping[str, object],
mm_hash_overrides: Optional[dict[str, list[str]]] = None,
) -> tuple[list[int], MultiModalProcessingInfo, bool]:
prompt_ids, mm_info, _ = super()._cached_apply_hf_processor(
prompt=prompt,
mm_data_items=mm_data_items,
hf_processor_mm_kwargs=hf_processor_mm_kwargs,
tokenization_kwargs=tokenization_kwargs,
mm_hash_overrides=mm_hash_overrides,
)
# NOTE: The tokens are already inserted by the chat template