biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Misc] Clean up type annotation for SupportsMultiModal (#14794)

Browse Source 
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung
2025-03-14 15:59:56 +08:00
committed by GitHub
parent 09269b3127
commit 601bd3268e
27 changed files with 121 additions and 141 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
vllm/model_executor/models/llava_onevision.py
View File

@@ -19,8 +19,7 @@ from vllm.model_executor.layers.activation import get_act_fn
from vllm.model_executor.layers.sampler import SamplerOutput, get_sampler
from vllm.model_executor.sampling_metadata import SamplingMetadata
from vllm.multimodal import MULTIMODAL_REGISTRY
from vllm.multimodal.inputs import (MultiModalFieldConfig, MultiModalKwargs,
NestedTensors)
from vllm.multimodal.inputs import MultiModalFieldConfig, MultiModalKwargs
from vllm.multimodal.parse import (ImageSize, MultiModalDataItems,
VideoEmbeddingItems, VideoProcessorItems)
from vllm.multimodal.processing import PromptReplacement, PromptUpdate
@@ -29,7 +28,7 @@ from vllm.sequence import IntermediateTensors
from vllm.utils import is_list_of
from .clip import CLIPVisionModel
from .interfaces import SupportsMultiModal, SupportsPP
from .interfaces import MultiModalEmbeddings, SupportsMultiModal, SupportsPP
from .llava import LlavaDummyInputsBuilder, init_vision_tower_for_llava
from .llava_next import (BaseLlavaNextMultiModalProcessor, LlavaNextLikeConfig,
LlavaNextProcessingInfo)
@@ -856,7 +855,7 @@ class LlavaOnevisionForConditionalGeneration(nn.Module, SupportsMultiModal,
return image_feature
def get_multimodal_embeddings(
self, **kwargs) -> Optional[tuple[torch.Tensor, ...]]:
self, **kwargs: object) -> Optional[MultiModalEmbeddings]:
modalities = self._parse_and_validate_multimodal_inputs(**kwargs)
if not modalities:
return None
@@ -882,7 +881,7 @@ class LlavaOnevisionForConditionalGeneration(nn.Module, SupportsMultiModal,
def get_input_embeddings(
self,
input_ids: torch.Tensor,
multimodal_embeddings: Optional[tuple[torch.Tensor, ...]] = None,
multimodal_embeddings: Optional[MultiModalEmbeddings] = None,
) -> torch.Tensor:
inputs_embeds = self.language_model.get_input_embeddings(input_ids)
if multimodal_embeddings is not None:
@@ -894,10 +893,9 @@ class LlavaOnevisionForConditionalGeneration(nn.Module, SupportsMultiModal,
def get_input_embeddings_v0(
self,
input_ids: torch.Tensor,
image_input: Optional[NestedTensors] = None,
video_input: Optional[NestedTensors] = None,
image_input: Optional[LlavaOnevisionImagePixelInputs] = None,
video_input: Optional[LlavaOnevisionVideoPixelInputs] = None,
) -> torch.Tensor:
inputs_embeds = self.get_input_embeddings(input_ids)
if image_input is not None:
image_embeds = self._process_image_input(image_input)