[Refactor] Consolidate sequence normalization and enc-dec parsing (#33928)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung
2026-02-06 15:43:47 +00:00
committed by GitHub
parent 4707f7ebb4
commit cd8b405bd0
38 changed files with 1271 additions and 863 deletions
+6 -7
View File
@@ -20,7 +20,7 @@ from typing import (
import numpy as np
from PIL.Image import Image
from typing_extensions import NotRequired, TypeVar
from typing_extensions import TypeVar
from vllm.utils.collection_utils import is_list_of
from vllm.utils.import_utils import LazyLoader
@@ -32,9 +32,13 @@ if TYPE_CHECKING:
import torch
import torch.types
from transformers.feature_extraction_utils import BatchFeature
from vllm.inputs.data import _InputOptions
else:
torch = LazyLoader("torch", globals(), "torch")
_InputOptions = dict
_T = TypeVar("_T")
HfImageItem: TypeAlias = Union["Image", np.ndarray, "torch.Tensor"]
@@ -1059,7 +1063,7 @@ A dictionary containing per-item placeholder ranges for each modality.
"""
class MultiModalInputs(TypedDict):
class MultiModalInputs(_InputOptions):
"""
Represents the outputs of
[`BaseMultiModalProcessor`][vllm.multimodal.processing.BaseMultiModalProcessor],
@@ -1084,11 +1088,6 @@ class MultiModalInputs(TypedDict):
`prompt_token_ids`.
"""
cache_salt: NotRequired[str]
"""
Optional cache salt to be used for prefix caching.
"""
class MultiModalEncDecInputs(MultiModalInputs):
"""