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
+16 -45
View File
@@ -30,7 +30,6 @@ 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.model_executor.models.interfaces import (
EagleModelMixin,
SupportsEagle,
@@ -40,7 +39,7 @@ from vllm.model_executor.models.interfaces import (
from vllm.model_executor.models.utils import (
AutoWeightsLoader,
PPMissingLayer,
is_pp_missing_parameter,
WeightsMapper,
make_empty_intermediate_tensors_factory,
make_layers,
maybe_prefix,
@@ -48,11 +47,6 @@ from vllm.model_executor.models.utils import (
from vllm.sequence import IntermediateTensors
from vllm.v1.attention.backend import AttentionType
STEP_PACKED_MODULES_MAPPING = {
"qkv_proj": ["q_proj", "k_proj", "v_proj"],
"gate_up_proj": ["gate_proj", "up_proj"],
}
def _get_step_alibi_slopes(total_num_heads: int) -> torch.Tensor:
"""Reference ALiBi slopes used by Step models."""
@@ -242,42 +236,6 @@ class StepDecoderLayer(nn.Module):
hidden_states = self.mlp(hidden_states)
return hidden_states, residual
def load_weights(self, weights: Iterable[tuple[str, torch.Tensor]]) -> set[str]:
stacked_params_mapping = [
(".qkv_proj", ".q_proj", "q"),
(".qkv_proj", ".k_proj", "k"),
(".qkv_proj", ".v_proj", "v"),
(".gate_up_proj", ".gate_proj", 0),
(".gate_up_proj", ".up_proj", 1),
]
params_dict = dict(self.named_parameters())
loaded_params: set[str] = set()
for name, loaded_weight in weights:
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) # type: ignore[name-defined]
weight_loader(param, loaded_weight)
loaded_params.add(name)
return loaded_params
class StepDecoderModel(nn.Module, EagleModelMixin):
def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
@@ -354,7 +312,20 @@ class StepDecoderModel(nn.Module, EagleModelMixin):
class Step1ForCausalLM(nn.Module, SupportsPP, SupportsEagle, SupportsEagle3):
packed_modules_mapping = STEP_PACKED_MODULES_MAPPING
hf_to_vllm_mapper = WeightsMapper(
orig_to_new_stacked={
# weight_name: (param_name, shard_id)
".q_proj": (".qkv_proj", "q"),
".k_proj": (".qkv_proj", "k"),
".v_proj": (".qkv_proj", "v"),
".gate_proj": (".gate_up_proj", 0),
".up_proj": (".gate_up_proj", 1),
}
)
packed_modules_mapping = {
"qkv_proj": ["q_proj", "k_proj", "v_proj"],
"gate_up_proj": ["gate_proj", "up_proj"],
}
def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
super().__init__()
@@ -413,4 +384,4 @@ class Step1ForCausalLM(nn.Module, SupportsPP, SupportsEagle, SupportsEagle3):
def load_weights(self, weights: Iterable[tuple[str, torch.Tensor]]) -> set[str]:
loader = AutoWeightsLoader(self)
return loader.load_weights(weights)
return loader.load_weights(weights, mapper=self.hf_to_vllm_mapper)