Remove unnecessary load_weights methods (#44589)

Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
This commit is contained in:
Harry Mellor
2026-06-29 01:52:23 -07:00
committed by GitHub
parent db28ae2d07
commit 5051698e41
54 changed files with 821 additions and 1975 deletions
+14 -38
View File
@@ -35,15 +35,14 @@ from vllm.model_executor.layers.vocab_parallel_embedding import (
ParallelLMHead,
VocabParallelEmbedding,
)
from vllm.model_executor.model_loader.weight_utils import default_weight_loader
from vllm.sequence import IntermediateTensors
from .interfaces import SupportsLoRA, SupportsPP
from .interfaces import SupportsLoRA, SupportsPP, SupportsQuant
from .interfaces_base import default_pooling_type
from .utils import (
AutoWeightsLoader,
StageMissingLayer,
is_pp_missing_parameter,
WeightsMapper,
make_empty_intermediate_tensors_factory,
make_layers,
maybe_prefix,
@@ -249,6 +248,14 @@ class InternLMDecoderLayer(nn.Module):
@support_torch_compile
class InternLM2Model(nn.Module):
hf_to_vllm_mapper = WeightsMapper(
orig_to_new_stacked={
# weight_name: (param_name, shard_id)
".w1": (".gate_up_proj", 0),
".w3": (".gate_up_proj", 1),
}
)
def __init__(
self,
*,
@@ -310,43 +317,12 @@ class InternLM2Model(nn.Module):
return hidden_states
def load_weights(self, weights: Iterable[tuple[str, torch.Tensor]]) -> set[str]:
stacked_params_mapping = [
# (param_name, shard_name, shard_id)
("gate_up_proj", "w1", 0),
("gate_up_proj", "w3", 1),
]
params_dict = dict(self.named_parameters())
loaded_params: set[str] = set()
for name, loaded_weight in weights:
if "rotary_emb.inv_freq" in name:
continue
for param_name, weight_name, shard_id in stacked_params_mapping:
if weight_name not in name:
continue
name = name.replace(weight_name, param_name)
# 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 = param.weight_loader
weight_loader(param, loaded_weight, shard_id)
break
else:
# 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, mapper=self.hf_to_vllm_mapper)
class InternLM2ForCausalLM(nn.Module, SupportsPP, SupportsLoRA):
class InternLM2ForCausalLM(nn.Module, SupportsPP, SupportsLoRA, SupportsQuant):
hf_to_vllm_mapper = InternLM2Model.hf_to_vllm_mapper
packed_modules_mapping = {
"wqkv": ["wqkv"],
"gate_up_proj": ["w1", "w3"],