[Misc][V1] Avoid using envs.VLLM_USE_V1 in mm processing (#14256)

Signed-off-by: Roger Wang <ywang@roblox.com>
This commit is contained in:
Roger Wang
2025-03-05 07:37:16 +00:00
committed by GitHub
parent 32985bed7c
commit ec79b67c77
7 changed files with 38 additions and 8 deletions
+3 -1
View File
@@ -175,8 +175,10 @@ class MllamaMultiModalProcessor(EncDecMultiModalProcessor[MllamaProcessingInfo]
prompt: Union[str, list[int]],
mm_data: MultiModalDataDict,
hf_processor_mm_kwargs: Mapping[str, object],
return_mm_hashes: bool = False,
) -> MultiModalEncDecInputs:
mm_inputs = super().apply(prompt, mm_data, hf_processor_mm_kwargs)
mm_inputs = super().apply(prompt, mm_data, hf_processor_mm_kwargs,
return_mm_hashes)
# Check that the number of image tokens in the decoder prompt matches
# the number of images provided in mm_data