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
+10 -67
View File
@@ -26,10 +26,6 @@ from vllm.model_executor.layers.vocab_parallel_embedding import (
ParallelLMHead,
VocabParallelEmbedding,
)
from vllm.model_executor.model_loader.weight_utils import (
default_weight_loader,
maybe_remap_kv_scale_name,
)
from vllm.sequence import IntermediateTensors
from .interfaces import (
@@ -42,7 +38,7 @@ from .interfaces import (
from .utils import (
AutoWeightsLoader,
PPMissingLayer,
is_pp_missing_parameter,
WeightsMapper,
make_empty_intermediate_tensors_factory,
make_layers,
maybe_prefix,
@@ -276,67 +272,6 @@ class ArceeModel(nn.Module, EagleModelMixin):
return hidden_states, aux_hidden_states
return hidden_states
def load_weights(self, weights: Iterable[tuple[str, torch.Tensor]]) -> set[str]:
"""Load weights, mapping q/k/v projections to fused qkv_proj."""
stacked_params_mapping = [
(".qkv_proj", ".q_proj", "q"),
(".qkv_proj", ".k_proj", "k"),
(".qkv_proj", ".v_proj", "v"),
]
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
if "rotary_emb.cos_cached" in name or "rotary_emb.sin_cached" in name:
continue
if "scale" in name or "zero_point" in name:
remapped_name = maybe_remap_kv_scale_name(name, params_dict)
if remapped_name is None:
continue
name = remapped_name
mapped = False
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)
if name.endswith(".bias") and name not in params_dict:
mapped = True
break
if is_pp_missing_parameter(name, self):
mapped = True
break
param = params_dict[name]
weight_loader = param.weight_loader # type: ignore[attr-defined]
weight_loader(param, loaded_weight, shard_id)
loaded_params.add(name)
mapped = True
break
if mapped:
continue
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 ArceeForCausalLM(
nn.Module, SupportsLoRA, SupportsPP, SupportsEagle, SupportsEagle3
@@ -344,6 +279,14 @@ class ArceeForCausalLM(
"""Arcee Model for causal language modeling, integrated with vLLM
runtime."""
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"),
}
)
# Map fused module names to their submodule components
# (for quantization and LoRA)
packed_modules_mapping = {
@@ -420,4 +363,4 @@ class ArceeForCausalLM(
)
# AutoWeightLoader handles weight name remapping, including fusing
# separate q_proj, k_proj, v_proj into qkv_proj
return loader.load_weights(weights)
return loader.load_weights(weights, mapper=self.hf_to_vllm_mapper)