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

[V0 Deprecation][Models] Remove all V0 condition for mm embeddings merge (#25331)

Browse Source 
Signed-off-by: Isotr0py <mozf@mail2.sysu.edu.cn>
Signed-off-by: isotr0py <2037008807@qq.com>
This commit is contained in:
Isotr0py
2025-09-29 14:09:18 +08:00
committed by GitHub
parent 65ecb4f134
commit bd51f78e39
42 changed files with 13 additions and 809 deletions
Download Patch File Download Diff File Expand all files Collapse all files

31
vllm/model_executor/models/qwen2_5_omni_thinker.py
View File

@@ -69,8 +69,7 @@ from vllm.utils.tensor_schema import TensorSchema, TensorShape
from .interfaces import (MultiModalEmbeddings, SupportsLoRA,
SupportsMultiModal, SupportsPP)
from .utils import (AutoWeightsLoader, WeightsMapper,
init_vllm_registered_model, maybe_prefix,
merge_multimodal_embeddings)
init_vllm_registered_model, maybe_prefix)
try:
import flash_attn
@@ -908,26 +907,6 @@ class Qwen2_5OmniThinkerForConditionalGeneration(
multimodal_embeddings.append((video_embeds, "video"))
return multimodal_embeddings
def get_input_embeddings_v0(
self,
input_ids: torch.Tensor,
multimodal_embeddings: Optional[NestedTensors] = None,
) -> torch.Tensor:
inputs_embeds = self.language_model.get_input_embeddings(input_ids)
if multimodal_embeddings is None or len(multimodal_embeddings) == 0:
return inputs_embeds
for embeddings, modality in multimodal_embeddings:
if modality == "audio":
placeholder_token_id = self.config.audio_token_index
if modality == "image":
placeholder_token_id = self.config.image_token_index
if modality == "video":
placeholder_token_id = self.config.video_token_index
inputs_embeds = merge_multimodal_embeddings(
input_ids, inputs_embeds, embeddings, placeholder_token_id)
return inputs_embeds
def forward(
self,
input_ids: torch.Tensor,
@@ -939,14 +918,6 @@ class Qwen2_5OmniThinkerForConditionalGeneration(
if intermediate_tensors is not None:
inputs_embeds = None
# NOTE: In v1, inputs_embeds is always generated at model runner, this
# condition is for v0 compatibility.
elif inputs_embeds is None:
multimodal_embeddings = self.get_multimodal_embeddings_v0(**kwargs)
inputs_embeds = self.get_input_embeddings_v0(
input_ids, multimodal_embeddings)
input_ids = None
hidden_states = self.language_model.model(input_ids,
positions,
intermediate_tensors,