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

42
vllm/model_executor/models/paligemma.py
View File

@@ -295,30 +295,32 @@ class PaliGemmaForConditionalGeneration(
multimodal_config = vllm_config.model_config.multimodal_config
self.config = config
self.multimodal_config = multimodal_config
self.vision_tower = SiglipVisionModel(
config.vision_config,
quant_config,
prefix=maybe_prefix(prefix, "vision_tower"),
)
self.multi_modal_projector = PaliGemmaMultiModalProjector(
vision_hidden_size=config.vision_config.hidden_size,
projection_dim=config.vision_config.projection_dim,
)
self.quant_config = quant_config
with self._mark_tower_model(vllm_config, "image"):
self.vision_tower = SiglipVisionModel(
config.vision_config,
quant_config,
prefix=maybe_prefix(prefix, "vision_tower"),
)
self.multi_modal_projector = PaliGemmaMultiModalProjector(
vision_hidden_size=config.vision_config.hidden_size,
projection_dim=config.vision_config.projection_dim,
)
if config.text_config.model_type == "gemma":
config.text_config.architectures = ["GemmaForCausalLM"]
else:
config.text_config.architectures = ["Gemma2ForCausalLM"]
self.language_model = init_vllm_registered_model(
vllm_config=vllm_config,
hf_config=config.text_config,
prefix=maybe_prefix(prefix, "language_model"),
)
logit_scale = getattr(config, "logit_scale", 1.0)
self.language_model.logits_processor.scale *= logit_scale
with self._mark_language_model(vllm_config):
self.language_model = language_model = init_vllm_registered_model(
vllm_config=vllm_config,
hf_config=config.text_config,
prefix=maybe_prefix(prefix, "language_model"),
)
logit_scale = getattr(config, "logit_scale", 1.0)
language_model.logits_processor.scale *= logit_scale
self.make_empty_intermediate_tensors = (
self.language_model.make_empty_intermediate_tensors
@@ -367,7 +369,6 @@ class PaliGemmaForConditionalGeneration(
if image_input["type"] == "image_embeds":
return image_input["data"]
assert self.vision_tower is not None
pixel_values = image_input["data"]
image_features = self._image_pixels_to_features(
self.vision_tower,
@@ -376,9 +377,6 @@ class PaliGemmaForConditionalGeneration(
return self.multi_modal_projector(image_features)
def get_language_model(self) -> torch.nn.Module:
return self.language_model
def embed_multimodal(self, **kwargs: object) -> MultiModalEmbeddings:
image_input = self._parse_and_validate_image_input(**kwargs)
if image_input is None: