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

[Model] Use context managers for encoder- and LM-only mode (#32605)

Browse Source 
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung
2026-01-20 11:43:38 +08:00
committed by GitHub
parent 6c01ffb897
commit 4753f3bf69
21 changed files with 290 additions and 353 deletions
Download Patch File Download Diff File Expand all files Collapse all files

35
vllm/model_executor/models/step3_vl.py
View File

@@ -942,7 +942,7 @@ class Step3VLForConditionalGeneration(nn.Module, SupportsMultiModal, SupportsPP)
self.multimodal_config = multimodal_config
self.use_data_parallel = multimodal_config.mm_encoder_tp_mode == "data"
if multimodal_config.get_limit_per_prompt("image"):
with self._mark_tower_model(vllm_config, "image"):
self.vision_model = Step3VisionTransformer(
config.vision_config,
None,
@@ -967,17 +967,13 @@ class Step3VLForConditionalGeneration(nn.Module, SupportsMultiModal, SupportsPP)
config.hidden_size,
bias=config.projector_bias,
)
else:
self.vision_model = None
self.vit_downsampler = None
self.vit_downsampler2 = None
self.vit_large_projector = None
self.language_model = init_vllm_registered_model(
vllm_config=vllm_config,
hf_config=config.text_config,
prefix=maybe_prefix(prefix, "language_model"),
)
with self._mark_language_model(vllm_config):
self.language_model = init_vllm_registered_model(
vllm_config=vllm_config,
hf_config=config.text_config,
prefix=maybe_prefix(prefix, "language_model"),
)
self.make_empty_intermediate_tensors = (
self.language_model.make_empty_intermediate_tensors
@@ -1071,9 +1067,6 @@ class Step3VLForConditionalGeneration(nn.Module, SupportsMultiModal, SupportsPP)
)
return merged_image_features
def get_language_model(self) -> torch.nn.Module:
return self.language_model
def embed_multimodal(self, **kwargs) -> MultiModalEmbeddings:
image_input = self._parse_and_validate_image_input(**kwargs)
if image_input is None:
@@ -1133,15 +1126,5 @@ class Step3VLForConditionalGeneration(nn.Module, SupportsMultiModal, SupportsPP)
return self.language_model.compute_logits(hidden_states)
def load_weights(self, weights: Iterable[tuple[str, torch.Tensor]]):
skip_prefixes = []
if self.vision_model is None and self.vit_large_projector is None:
skip_prefixes = [
"vision_model.",
"vit_downsampler.",
"vit_downsampler2.",
"vit_large_projector.",
]
loader = AutoWeightsLoader(self, skip_prefixes=skip_prefixes)
loaded_weights = loader.load_weights(weights, mapper=self.hf_to_vllm_mapper)
return loaded_weights
loader = AutoWeightsLoader(self)
return loader.load_weights(weights, mapper=self.hf_to_vllm_mapper)