[Refactor] Decouple TimingContext from InputProcessingContext (#35083)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung
2026-02-23 14:15:50 +00:00
committed by GitHub
parent 1e8438a89a
commit 392645454b
38 changed files with 419 additions and 649 deletions
+7 -20
View File
@@ -50,16 +50,17 @@ from vllm.multimodal.parse import (
ImageProcessorItems,
ImageSize,
MultiModalDataItems,
MultiModalUUIDItems,
)
from vllm.multimodal.processing import BaseDummyInputsBuilder, ProcessorInputs
from vllm.multimodal.processing import BaseDummyInputsBuilder
from vllm.multimodal.processing.processor import (
BaseMultiModalProcessor,
BaseProcessingInfo,
MultiModalProcessingInfo,
ProcessorInputs,
PromptReplacement,
PromptUpdate,
PromptUpdateDetails,
TimingContext,
)
from vllm.platforms import current_platform
from vllm.sequence import IntermediateTensors
@@ -277,7 +278,6 @@ class PixtralDummyInputsBuilder(BaseDummyInputsBuilder[PixtralProcessingInfo]):
dummy_text = self.get_dummy_text(mm_counts)
dummy_mm_data = self.get_dummy_mm_data(seq_len, mm_counts, mm_options)
dummy_images = dummy_mm_data.get("image", [])
tokenization_kwargs = {"truncation": False}
request = ChatCompletionRequest(
messages=[
@@ -294,11 +294,7 @@ class PixtralDummyInputsBuilder(BaseDummyInputsBuilder[PixtralProcessingInfo]):
dummy_mm_items = self.info.parse_mm_data(dummy_mm_data)
return ProcessorInputs(
prompt=dummy_tokens,
mm_items=dummy_mm_items,
tokenization_kwargs=tokenization_kwargs,
)
return ProcessorInputs(prompt=dummy_tokens, mm_data_items=dummy_mm_items)
class PixtralMultiModalProcessor(BaseMultiModalProcessor[PixtralProcessingInfo]):
@@ -344,19 +340,10 @@ class PixtralMultiModalProcessor(BaseMultiModalProcessor[PixtralProcessingInfo])
def _cached_apply_hf_processor(
self,
prompt: str | list[int],
mm_data_items: MultiModalDataItems,
mm_uuid_items: MultiModalUUIDItems | None,
hf_processor_mm_kwargs: Mapping[str, object],
tokenization_kwargs: Mapping[str, object],
inputs: ProcessorInputs,
timing_ctx: TimingContext,
) -> tuple[list[int], MultiModalProcessingInfo, bool]:
prompt_ids, mm_info, _ = super()._cached_apply_hf_processor(
prompt=prompt,
mm_data_items=mm_data_items,
mm_uuid_items=mm_uuid_items,
hf_processor_mm_kwargs=hf_processor_mm_kwargs,
tokenization_kwargs=tokenization_kwargs,
)
prompt_ids, mm_info, _ = super()._cached_apply_hf_processor(inputs, timing_ctx)
# NOTE: The tokens are already inserted by the chat template
return prompt_ids, mm_info, True