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

[4/N] Initialize MM components in context managers (M-P) (#32663)

Browse Source 
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung
2026-01-20 22:06:32 +08:00
committed by GitHub
parent bb9172030e
commit fda3f03eb2
24 changed files with 290 additions and 353 deletions
Download Patch File Download Diff File Expand all files Collapse all files

26
vllm/model_executor/models/molmo2.py
View File

@@ -2514,16 +2514,19 @@ class Molmo2ForConditionalGeneration(
kwargs[field.name] = getattr(config.adapter_config, field.name)
adapter_config = AdapterConfig(**kwargs)
self.vision_backbone = Molmo2VisionBackbone(
vit_config,
adapter_config,
quant_config,
prefix=maybe_prefix(prefix, "vision_backbone"),
)
self.model = Molmo2TextModel(
vllm_config=vllm_config,
prefix=maybe_prefix(prefix, "model"),
)
with self._mark_tower_model(vllm_config, {"image", "video"}):
self.vision_backbone = Molmo2VisionBackbone(
vit_config,
adapter_config,
quant_config,
prefix=maybe_prefix(prefix, "vision_backbone"),
)
with self._mark_language_model(vllm_config):
self.model = Molmo2TextModel(
vllm_config=vllm_config,
prefix=maybe_prefix(prefix, "model"),
)
self.img_patch_id = config.image_patch_id
@@ -2687,9 +2690,6 @@ class Molmo2ForConditionalGeneration(
out.append(out_features)
return tuple(out)
def get_language_model(self) -> torch.nn.Module:
return self.model
def embed_multimodal(self, **kwargs: object) -> MultiModalEmbeddings | None:
modalities = self._parse_and_validate_multimodal_inputs(**kwargs)
if not modalities: