[Refactor] Define MultiModalKwargsItems separate from MultiModalKwargs (#23053)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung
2025-08-18 09:52:00 +00:00
committed by GitHub
parent 5c79b0d648
commit 27e8d1ea3e
77 changed files with 431 additions and 383 deletions
+4 -3
View File
@@ -56,7 +56,8 @@ from vllm.model_executor.models.module_mapping import MultiModelKeys
from vllm.model_executor.sampling_metadata import SamplingMetadata
from vllm.multimodal import MULTIMODAL_REGISTRY
from vllm.multimodal.inputs import (MultiModalDataDict, MultiModalEncDecInputs,
MultiModalFieldConfig, MultiModalKwargs)
MultiModalFieldConfig,
MultiModalKwargsItems)
from vllm.multimodal.parse import (ImageProcessorItems, ImageSize,
MultiModalDataItems)
from vllm.multimodal.processing import (BaseProcessingInfo,
@@ -217,7 +218,7 @@ class MllamaMultiModalProcessor(EncDecMultiModalProcessor[MllamaProcessingInfo]
# Set encoder prompt length based on the number of tiles.
# This tells the block manager to allocate correct number
# of slots for encoder tokens.
num_tiles = mm_inputs["mm_kwargs"]["num_tiles"]
num_tiles = mm_inputs["mm_kwargs"].get_data()["num_tiles"]
decode_tiles = num_tiles[num_encode_images:num_images].sum().item()
num_tokens = decode_tiles * token_per_chunk
mm_inputs["encoder_prompt_token_ids"] = [image_token_id
@@ -302,7 +303,7 @@ class MllamaMultiModalProcessor(EncDecMultiModalProcessor[MllamaProcessingInfo]
self,
mm_items: MultiModalDataItems,
hf_processor_mm_kwargs: Mapping[str, object],
out_mm_kwargs: MultiModalKwargs,
out_mm_kwargs: MultiModalKwargsItems,
) -> Sequence[PromptUpdate]:
token_per_chunk = self.info.get_token_per_chunk_from_config()
image_token_id = self.info.get_hf_config().image_token_index