[VLM] Generalized prompt updates for multi-modal processor (#13964)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung
2025-02-27 17:44:25 +00:00
committed by GitHub
parent 7864875879
commit f1579b229d
29 changed files with 629 additions and 486 deletions
+10 -14
View File
@@ -1,9 +1,10 @@
# SPDX-License-Identifier: Apache-2.0
import math
from collections import OrderedDict
from collections.abc import Iterable, Mapping, Sequence
from functools import cached_property
from typing import (Iterable, List, Literal, Mapping, Optional, OrderedDict,
Set, Tuple, TypedDict, Union)
from typing import List, Literal, Optional, Set, Tuple, TypedDict, Union
import torch
import torch.nn as nn
@@ -24,8 +25,7 @@ from vllm.multimodal.inputs import MultiModalFieldConfig, MultiModalKwargs
from vllm.multimodal.parse import MultiModalDataDict, MultiModalDataItems
from vllm.multimodal.processing import (BaseProcessingInfo,
EncDecMultiModalProcessor,
PromptReplacement,
PromptReplacementDetails)
PromptInsertion, PromptUpdate)
from vllm.multimodal.profiling import BaseDummyInputsBuilder, ProcessorInputs
from vllm.sequence import IntermediateTensors
@@ -803,7 +803,7 @@ class Florence2DummyInputsBuilder(
class Florence2MultiModalProcessor(
EncDecMultiModalProcessor[Florence2ProcessingInfo]):
def _hf_processor_applies_repl(
def _hf_processor_applies_updates(
self,
prompt_text: str,
mm_items: MultiModalDataItems,
@@ -850,26 +850,22 @@ class Florence2MultiModalProcessor(
) -> Mapping[str, MultiModalFieldConfig]:
return dict(pixel_values=MultiModalFieldConfig.batched("image"))
def _get_prompt_replacements(
def _get_prompt_updates(
self,
mm_items: MultiModalDataItems,
hf_processor_mm_kwargs: Mapping[str, object],
out_mm_kwargs: MultiModalKwargs,
) -> list[PromptReplacement]:
) -> Sequence[PromptUpdate]:
hf_config = self.info.get_hf_config()
pad_token_id = hf_config.pad_token_id
bos_token_id = hf_config.bos_token_id
num_image_tokens = self.info.get_max_image_tokens()
image_tokens = [pad_token_id] * num_image_tokens
return [
PromptReplacement(
PromptInsertion(
modality="image",
target=[bos_token_id],
replacement=PromptReplacementDetails(
full=image_tokens + [bos_token_id],
features=image_tokens,
),
target="",
insertion=image_tokens,
)
]