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

[V1]SupportsV0Only protocol for model definitions (#13959)

Browse Source 
Signed-off-by: Roger Wang <ywang@roblox.com>
This commit is contained in:
Roger Wang
2025-02-27 17:02:15 -08:00
committed by GitHub
parent 67fc426845
commit 6c85da3a18
19 changed files with 93 additions and 32 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
vllm/model_executor/models/paligemma.py
View File

@@ -18,7 +18,7 @@ from vllm.multimodal.inputs import NestedTensors
from vllm.sequence import IntermediateTensors
from vllm.transformers_utils.tokenizer import cached_tokenizer_from_config
from .interfaces import SupportsMultiModal, SupportsPP
from .interfaces import SupportsMultiModal, SupportsPP, SupportsV0Only
from .siglip import (SiglipVisionModel, dummy_image_for_siglip,
dummy_seq_data_for_siglip, get_max_siglip_image_tokens)
from .utils import (AutoWeightsLoader, init_vllm_registered_model,
@@ -136,7 +136,7 @@ class PaliGemmaMultiModalProjector(nn.Module):
@INPUT_REGISTRY.register_dummy_data(dummy_data_for_paligemma)
@INPUT_REGISTRY.register_input_processor(input_processor_for_paligemma)
class PaliGemmaForConditionalGeneration(nn.Module, SupportsMultiModal,
SupportsPP):
SupportsPP, SupportsV0Only):
packed_modules_mapping = {
"qkv_proj": [
"q_proj",