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
+17 -32
View File
@@ -30,7 +30,6 @@ from vllm.config.multimodal import BaseDummyOptions
from vllm.inputs import MultiModalDataDict
from vllm.model_executor.layers.attention.mm_encoder_attention import MMEncoderAttention
from vllm.model_executor.layers.linear import QKVParallelLinear, RowParallelLinear
from vllm.model_executor.model_loader.weight_utils import default_weight_loader
from vllm.multimodal import MULTIMODAL_REGISTRY
from vllm.multimodal.inputs import (
MultiModalFieldConfig,
@@ -53,7 +52,12 @@ from vllm.sequence import IntermediateTensors
from vllm.utils.import_utils import _has_module
from .interfaces import MultiModalEmbeddings, SupportsMultiModal, SupportsPP
from .utils import AutoWeightsLoader, init_vllm_registered_model, maybe_prefix
from .utils import (
AutoWeightsLoader,
WeightsMapper,
init_vllm_registered_model,
maybe_prefix,
)
class _SinusoidsPositionEmbedding(nn.Module):
@@ -79,6 +83,14 @@ class _SinusoidsPositionEmbedding(nn.Module):
class FunAudioChatAudioAttention(nn.Module):
"""Multi-headed attention used inside the continuous audio tower."""
hf_to_vllm_mapper = WeightsMapper(
orig_to_new_stacked={
".q_proj": (".qkv_proj", "q"),
".k_proj": (".qkv_proj", "k"),
".v_proj": (".qkv_proj", "v"),
}
)
def __init__(self, config: Any):
super().__init__()
self.embed_dim = int(config.d_model)
@@ -123,42 +135,15 @@ class FunAudioChatAudioAttention(nn.Module):
bias=True,
)
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"),
]
params_dict = dict(self.named_parameters())
with torch.no_grad():
if self.qkv_proj.bias is not None:
# HF FunAudioChat uses bias=False for k_proj. Ensure the missing
# shard starts as zeros, while allowing q/v shards to load.
self.qkv_proj.bias.zero_()
loaded_params: set[str] = set()
for name, loaded_weight in weights:
for param_name, shard_name, shard_id in stacked_params_mapping:
if shard_name not in name:
continue
name = name.replace(shard_name, param_name)
param = params_dict[name]
weight_loader = getattr(param, "weight_loader", default_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
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
def load_weights(self, weights: Iterable[tuple[str, torch.Tensor]]) -> set[str]:
loader = AutoWeightsLoader(self)
return loader.load_weights(weights, mapper=self.hf_to_vllm_mapper)
def forward(
self,