[Refactor] Get prompt updates earlier (#23097)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung
2025-08-18 12:31:53 +00:00
committed by GitHub
parent 5a30bd10d8
commit d3f71f1224
6 changed files with 84 additions and 69 deletions
+5 -6
View File
@@ -35,7 +35,8 @@ from vllm.multimodal.inputs import (MultiModalDataDict, MultiModalFieldConfig,
from vllm.multimodal.parse import (AudioProcessorItems, MultiModalDataItems,
MultiModalDataParser)
from vllm.multimodal.processing import (BaseMultiModalProcessor,
BaseProcessingInfo, MultiModalHashes,
BaseProcessingInfo,
MultiModalProcessingInfo,
PromptReplacement, PromptUpdate)
from vllm.multimodal.profiling import BaseDummyInputsBuilder, ProcessorInputs
from vllm.sequence import IntermediateTensors
@@ -289,10 +290,8 @@ class VoxtralMultiModalProcessor(BaseMultiModalProcessor[VoxtralProcessingInfo]
tokenization_kwargs: Mapping[str, object],
*,
return_mm_hashes: bool,
) -> tuple[list[int], MultiModalKwargsItems, Optional[MultiModalHashes],
bool]:
prompt_ids, mm_kwargs, mm_hashes, _ = super(
)._cached_apply_hf_processor(
) -> 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,
@@ -301,7 +300,7 @@ class VoxtralMultiModalProcessor(BaseMultiModalProcessor[VoxtralProcessingInfo]
)
# NOTE: The tokens are already inserted by the chat template
return prompt_ids, mm_kwargs, mm_hashes, True
return prompt_ids, mm_info, True
def _get_data_parser(self) -> MultiModalDataParser:
sampling_rate = self.info.get_hf_processor().sampling_rate