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 15, 12:15 UTC
[Bugfix] Add missing `get_language_model` to new MLLMs (#17300)
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
Cyrus Leung committed
Apr 28, 2025 at 19:09 UTC
344e193b7ddb654c95d1195278edd9e129717979
2 files changed
+6
vllm/model_executor/models/kimi_vl.py
+3
@@ -369,6 +369,9 @@ class KimiVLForConditionalGeneration(nn.Module, SupportsMultiModal):
369
return self.multi_modal_projector(
370
torch.cat(image_features)).split(lengths)
371
372
+ def get_language_model(self) -> torch.nn.Module:
373
+ return self.language_model
374
+
375
def get_multimodal_embeddings(self,
376
**kwargs: object) -> Optional[NestedTensors]:
377
# Validate the multimodal input keyword arguments
vllm/model_executor/models/qwen2_5_omni_thinker.py
+3
@@ -809,6 +809,9 @@ class Qwen2_5OmniThinkerForConditionalGeneration(
809
"audio"] = self._parse_and_validate_audio_input(**kwargs)
810
return mm_input_by_modality
811
812
+ def get_language_model(self) -> torch.nn.Module:
813
+ return self.language_model
814
+
815
def get_multimodal_embeddings(
816
self, **kwargs: object) -> Optional[MultiModalEmbeddings]:
817