Remove even more unnecessary load_weights methods (#48496)

Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
This commit is contained in:
Harry Mellor
2026-07-18 08:40:27 +00:00
committed by GitHub
parent d96aee0951
commit c233d90aa8
44 changed files with 636 additions and 2073 deletions
+26 -43
View File
@@ -31,10 +31,6 @@ 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.multimodal.inputs import NestedTensors
from vllm.transformers_utils.config import set_default_rope_theta
from vllm.transformers_utils.repo_utils import get_hf_file_bytes
@@ -44,6 +40,7 @@ from .qwen2 import Qwen2MLP as Qwen3MLP
from .qwen3 import Qwen3ForCausalLM
from .utils import (
AutoWeightsLoader,
WeightsMapper,
get_draft_quant_config,
maybe_prefix,
process_eagle_weight,
@@ -333,6 +330,17 @@ class DFlashQwen3DecoderLayer(nn.Module):
@support_torch_compile
class DFlashQwen3Model(nn.Module):
hf_to_vllm_mapper = WeightsMapper(
orig_to_new_substr={"midlayer.": "layers.0."},
orig_to_new_stacked={
".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),
},
)
def __init__(
self,
*,
@@ -624,51 +632,26 @@ class DFlashQwen3Model(nn.Module):
hidden_states, _ = self.norm(hidden_states, residual)
return hidden_states
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()
tp_rank = get_tensor_model_parallel_rank()
def _preprocess(
self, weights: Iterable[tuple[str, torch.Tensor]]
) -> Iterable[tuple[str, torch.Tensor]]:
tp_size = get_tensor_model_parallel_world_size()
tp_rank = get_tensor_model_parallel_rank()
for name, loaded_weight in weights:
if "midlayer." in name:
name = name.replace("midlayer.", "layers.0.")
if "scale" in name:
name = maybe_remap_kv_scale_name(name, params_dict)
if name is None:
continue
if "attention_sink_bias" in name:
if name not in params_dict:
continue
# Sink bias is per-head; shard it across TP ranks like the
# attention heads themselves.
param = params_dict[name]
heads_per_rank = loaded_weight.shape[0] // tp_size
head_start = tp_rank * heads_per_rank
narrow_weight = loaded_weight.narrow(0, head_start, heads_per_rank)
param.data.copy_(narrow_weight)
loaded_params.add(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)
param = params_dict[name]
weight_loader = param.weight_loader
weight_loader(param, loaded_weight, shard_id)
break
else:
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
loaded_weight = loaded_weight.narrow(
0, tp_rank * heads_per_rank, heads_per_rank
)
yield name, loaded_weight
def load_weights(self, weights: Iterable[tuple[str, torch.Tensor]]) -> set[str]:
loader = AutoWeightsLoader(self)
return loader.load_weights(
self._preprocess(weights), mapper=self.hf_to_vllm_mapper
)
class DFlashQwen3ForCausalLM(Qwen3ForCausalLM):