A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 11, 06:15 UTC
[Chore] Remove v0 dead code for Qwen2.5-omni (#30883)
Signed-off-by: Isotr0py <mozf@mail2.sysu.edu.cn>
Isotr0py committed
Dec 18, 2025 at 11:54 UTC
6fe58876528751a808c0d25016d3c97b432b909f
1 file changed
-22
vllm/model_executor/models/qwen2_5_omni_thinker.py
-22
@@ -70,7 +70,6 @@ from vllm.multimodal.inputs import (
70
MultiModalFeatureSpec,
71
MultiModalFieldConfig,
72
MultiModalKwargsItems,
73
- NestedTensors,
73
)
74
from vllm.multimodal.parse import (
75
AudioProcessorItems,
@@ -1150,27 +1149,6 @@ class Qwen2_5OmniThinkerForConditionalGeneration(
1149
handle_oov_mm_token=handle_oov_mm_token,
1150
)
1151
1153
- def embed_multimodal_v0(self, **kwargs: object) -> NestedTensors | None:
1154
- audio_input = self._parse_and_validate_audio_input(**kwargs)
1155
- image_input = self._parse_and_validate_image_input(**kwargs)
1156
- video_input = self._parse_and_validate_video_input(**kwargs)
1157
-
1158
- if audio_input is None and image_input is None and video_input is None:
1159
- return None
1160
-
1161
- multimodal_embeddings: list[tuple[NestedTensors, str]] = []
1162
-
1163
- if audio_input is not None:
1164
- audio_embeds = self._process_audio_input(audio_input)
1165
- multimodal_embeddings.append((audio_embeds, "audio"))
1166
- if image_input is not None:
1167
- image_embeds = self._process_image_input(image_input)
1168
- multimodal_embeddings.append((image_embeds, "image"))
1169
- if video_input is not None:
1170
- video_embeds = self._process_video_input(video_input)
1171
- multimodal_embeddings.append((video_embeds, "video"))
1172
- return multimodal_embeddings
1173
-
1152
def forward(
1153
self,
1154
input_ids: torch.Tensor,