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

[Model] Use autoweightloader for mamba (#16950)

Browse Source 
Signed-off-by: sfeng33 <4florafeng@gmail.com>
This commit is contained in:
Flora Feng
2025-04-22 00:48:15 -07:00
committed by GitHub
parent acba33a0f1
commit 2689d5c027
1 changed files with 23 additions and 18 deletions
Download Patch File Download Diff File Expand all files Collapse all files

41
vllm/model_executor/models/mamba.py
View File

@ -27,7 +27,7 @@ from vllm.model_executor.sampling_metadata import SamplingMetadata
from vllm.sequence import IntermediateTensors
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,
maybe_prefix)
@ -154,6 +154,26 @@ class MambaModel(nn.Module):
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 MambaForCausalLM(nn.Module, HasInnerState, IsAttentionFree, SupportsPP,
SupportsV0Only):
@ -257,20 +277,5 @@ class MambaForCausalLM(nn.Module, HasInnerState, IsAttentionFree, SupportsPP,
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
loader = AutoWeightsLoader(self)
return loader.load_weights(weights)