xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-14 06:15:01 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Quant] Molmo SupportsQuant (#13336)

Browse Source
This commit is contained in:
Kyle Sayers 2025-02-18 00:34:47 -05:00 committed by GitHub
parent 932b51cedd
commit 88787bce1d
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 7 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
vllm/model_executor/models/molmo.py
View File

@ -52,7 +52,8 @@ from vllm.multimodal.profiling import BaseDummyInputsBuilder, ProcessorInputs
from vllm.sequence import IntermediateTensors from vllm.sequence import IntermediateTensors
from vllm.utils import JSONTree, json_map_leaves from vllm.utils import JSONTree, json_map_leaves
from .interfaces import SupportsLoRA, SupportsMultiModal, SupportsPP from .interfaces import (SupportsLoRA, SupportsMultiModal, SupportsPP,
SupportsQuant)
from .utils import (AutoWeightsLoader, WeightsMapper, flatten_bn, from .utils import (AutoWeightsLoader, WeightsMapper, flatten_bn,
is_pp_missing_parameter, is_pp_missing_parameter,
make_empty_intermediate_tensors_factory, make_layers, make_empty_intermediate_tensors_factory, make_layers,
@ -633,7 +634,8 @@ class MolmoDecoderNormAfterLayer(MolmoDecoderLayer):
return hidden_states, residual return hidden_states, residual
class MolmoVisionBackbone(nn.Module): class MolmoVisionBackbone(nn.Module, SupportsQuant):
packed_modules_mapping = {"merged_linear": ["gate_proj", "up_proj"]}
def __init__( def __init__(
self, self,
@ -794,7 +796,7 @@ class MolmoVisionBackbone(nn.Module):
@support_torch_compile @support_torch_compile
class MolmoModel(nn.Module): class MolmoModel(nn.Module, SupportsQuant):
def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""): def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
super().__init__() super().__init__()
@ -1402,8 +1404,8 @@ class MolmoMultiModalProcessor(BaseMultiModalProcessor[MolmoProcessingInfo]):
@MULTIMODAL_REGISTRY.register_processor(MolmoMultiModalProcessor, @MULTIMODAL_REGISTRY.register_processor(MolmoMultiModalProcessor,
info=MolmoProcessingInfo, info=MolmoProcessingInfo,
dummy_inputs=MolmoDummyInputsBuilder) dummy_inputs=MolmoDummyInputsBuilder)
class MolmoForCausalLM(nn.Module, SupportsMultiModal, SupportsPP, class MolmoForCausalLM(nn.Module, SupportsMultiModal, SupportsPP, SupportsLoRA,
SupportsLoRA): SupportsQuant):
hf_to_vllm_mapper = WeightsMapper( hf_to_vllm_mapper = WeightsMapper(
orig_to_new_substr={ orig_to_new_substr={
# vision backbone mapping # vision backbone mapping