Remove even more unnecessary load_weights methods (#48496)
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
This commit is contained in:
@@ -65,7 +65,12 @@ from .interfaces import (
|
||||
SupportsPP,
|
||||
SupportsTranscription,
|
||||
)
|
||||
from .utils import AutoWeightsLoader, init_vllm_registered_model, maybe_prefix
|
||||
from .utils import (
|
||||
AutoWeightsLoader,
|
||||
WeightsMapper,
|
||||
init_vllm_registered_model,
|
||||
maybe_prefix,
|
||||
)
|
||||
from .whisper import ISO639_1_SUPPORTED_LANGS, _create_fake_bias_for_k_proj
|
||||
|
||||
|
||||
@@ -395,6 +400,14 @@ class GlmAsrEncoder(nn.Module):
|
||||
"qkv_proj": ["q_proj", "k_proj", "v_proj"],
|
||||
}
|
||||
|
||||
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,
|
||||
@@ -496,44 +509,9 @@ class GlmAsrEncoder(nn.Module):
|
||||
return _GlmAsrEncoderOutput(last_hidden_state=hidden_states)
|
||||
|
||||
def load_weights(self, weights: Iterable[tuple[str, torch.Tensor]]) -> set[str]:
|
||||
"""Custom weight loading to handle q_proj/k_proj/v_proj -> qkv_proj mapping."""
|
||||
from vllm.model_executor.model_loader.weight_utils import default_weight_loader
|
||||
|
||||
weights = _create_fake_bias_for_k_proj(weights, ".k_proj.weight")
|
||||
|
||||
stacked_params_mapping = [
|
||||
# (param_name, shard_name, shard_id)
|
||||
("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:
|
||||
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
|
||||
|
||||
param = params_dict[name]
|
||||
weight_loader = param.weight_loader
|
||||
weight_loader(param, loaded_weight, shard_id)
|
||||
break
|
||||
else:
|
||||
# Default weight loading for non-stacked params
|
||||
if name.endswith(".bias") and name not in params_dict:
|
||||
continue
|
||||
if 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
|
||||
loader = AutoWeightsLoader(self)
|
||||
return loader.load_weights(weights, mapper=self.hf_to_vllm_mapper)
|
||||
|
||||
|
||||
class GlmAsrFeatureInputs(TensorSchema):
|
||||
|
||||
Reference in New Issue
Block a user