Skip to content

Commit

Permalink
[Misc] Skip loading extra bias for Qwen2-MOE GPTQ models (vllm-projec…
Browse files Browse the repository at this point in the history
  • Loading branch information
jeejeelee authored Sep 11, 2024
1 parent 94144e7 commit e497b8a
Showing 1 changed file with 8 additions and 2 deletions.
10 changes: 8 additions & 2 deletions vllm/model_executor/models/qwen2_moe.py
Original file line number Diff line number Diff line change
Expand Up @@ -469,7 +469,8 @@ def load_weights(self, weights: Iterable[Tuple[str, torch.Tensor]]):
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:
if ((name.endswith(".bias") or name.endswith("_bias"))
and name not in params_dict):
continue
# Skip layers on other devices.
if is_pp_missing_parameter(name, self):
Expand All @@ -490,6 +491,10 @@ def load_weights(self, weights: Iterable[Tuple[str, torch.Tensor]]):
# Skip layers on other devices.
if is_pp_missing_parameter(name, self):
continue
# Skip loading extra bias for GPTQ models.
if ((name.endswith(".bias") or name.endswith("_bias"))
and name not in params_dict):
continue
param = params_dict[name]
weight_loader = param.weight_loader
weight_loader(param,
Expand All @@ -500,7 +505,8 @@ def load_weights(self, weights: Iterable[Tuple[str, torch.Tensor]]):
break
else:
# Skip loading extra bias for GPTQ models.
if name.endswith(".bias") and name not in params_dict:
if ((name.endswith(".bias") or name.endswith("_bias"))
and name not in params_dict):
continue
# Skip layers on other devices.
if is_pp_missing_parameter(name, self):
Expand Down

0 comments on commit e497b8a

Please sign in to comment.