xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-10 07:15:01 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Model] Use AutoWeightsLoader for mamba2 (#18918)

Browse Source 
Signed-off-by: iLeGend <824040212@qq.com>
This commit is contained in:
iLeGend 2025-05-30 12:50:10 +08:00 committed by GitHub
parent 77164dad5e
commit 3987e2ae96
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 24 additions and 19 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

43
vllm/model_executor/models/mamba2.py
View File

@ -32,7 +32,7 @@ from vllm.model_executor.sampling_metadata import SamplingMetadata
from vllm.sequence import IntermediateTensors from vllm.sequence import IntermediateTensors
from vllm.utils import LayerBlockType from vllm.utils import LayerBlockType
from .utils import (is_pp_missing_parameter, from .utils import (AutoWeightsLoader, is_pp_missing_parameter,
make_empty_intermediate_tensors_factory, make_layers, make_empty_intermediate_tensors_factory, make_layers,
maybe_prefix) maybe_prefix)
@ -167,6 +167,27 @@ class Mamba2Model(nn.Module):
return hidden_states return hidden_states
def load_weights(self, weights: Iterable[tuple[str,
torch.Tensor]]) -> set[str]:
params_dict = dict(self.named_parameters())
loaded_params: set[str] = set()
for name, loaded_weight in weights:
if "A_log" in name:
name = name.replace("A_log", "A")
# Skip loading extra bias for GPTQ models.
if name.endswith(".bias") and name not in params_dict:
continue
if is_pp_missing_parameter(name, self):
continue
param = params_dict[name]
weight_loader = getattr(param, "weight_loader",
default_weight_loader)
weight_loader(param, loaded_weight)
loaded_params.add(name)
return loaded_params
class Mamba2ForCausalLM(nn.Module, HasInnerState, IsAttentionFree, class Mamba2ForCausalLM(nn.Module, HasInnerState, IsAttentionFree,
SupportsV0Only): SupportsV0Only):
@ -282,21 +303,5 @@ class Mamba2ForCausalLM(nn.Module, HasInnerState, IsAttentionFree,
def load_weights(self, weights: Iterable[tuple[str, def load_weights(self, weights: Iterable[tuple[str,
torch.Tensor]]) -> set[str]: torch.Tensor]]) -> set[str]:
params_dict = dict(self.named_parameters()) loader = AutoWeightsLoader(self)
loaded_params: set[str] = set() return loader.load_weights(weights)
for name, loaded_weight in weights:
if "A_log" in name:
name = name.replace("A_log", "A")
# Skip loading extra bias for GPTQ models.
if name.endswith(".bias") and name not in params_dict:
continue
if is_pp_missing_parameter(name, self):
continue
param = params_dict[name]
weight_loader = getattr(param, "weight_loader",
default_weight_loader)
weight_loader(param, loaded_weight)
loaded_params.add(name)
return loaded_params