Skip to content

fix bug of lwq gtpq #2128

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 4 commits into from
Feb 25, 2025
+8 −4
Merged

fix bug of lwq gtpq #2128

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
4c45d10
fix bug
n1ck-guo Feb 24, 2025
63bea6b
update
n1ck-guo Feb 24, 2025
15c39f1
fix
n1ck-guo Feb 25, 2025
d6b8524
[pre-commit.ci] auto fixes from pre-commit.com hooks
pre-commit-ci[bot] Feb 25, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
8 changes: 6 additions & 2 deletions neural_compressor/adaptor/torch_utils/gptq.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -718,10 +718,12 @@ def tmp(_, inp, out):
for n, p in sub_layer.named_parameters():
param_name = full_layer_name + "." + n
if n == "weight":
set_module_tensor_to_device(self.model, param_name, self.device, Q)
set_module_tensor_to_device(self.model, param_name, self.device, Q, dtype=Q.dtype)
else:
value = load_value(self.model, param_name, model_path)
set_module_tensor_to_device(self.model, param_name, self.device, value)
set_module_tensor_to_device(
self.model, param_name, self.device, value, dtype=value.dtype
)
# sub_layer.weight.data = Q
torch.save(sub_layer.state_dict(), LWQ_WORKSPACE + f"/{full_layer_name}.pt")
clean_module_weight(sub_layer)
Expand All @@ -745,6 +747,8 @@ def tmp(_, inp, out):
for j in range(len(self.dataloader)):
cache_keyword_batch = self.gather_single_batch_from_dict(self.cache_key_arguments, j)
cache_positional_batch = self.gather_single_batch_from_list(self.cache_positional_arguments, j)
# breakpoint()
# transformer_block = transformer_block.to(getattr(torch, self.model.config.torch_dtype))
out = transformer_block(*cache_positional_batch, **cache_keyword_batch)
out = self.track_hidden_states(out)
outs.append(out)
Expand Down
4 changes: 2 additions & 2 deletions neural_compressor/adaptor/torch_utils/layer_wise_quant/utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -221,7 +221,7 @@ def load_module(model, module_name, path, device="cpu"):
for n, p in module.named_parameters():
param_name = module_name + "." + n
value = load_value(model, param_name, path)
set_module_tensor_to_device(model, param_name, device, value)
set_module_tensor_to_device(model, param_name, device, value, dtype=value.dtype)


def register_weight_hooks(model, path, device="cpu", clean_weight=True, saved_path=None):
Expand All @@ -239,7 +239,7 @@ def hook(module, input):
value = state_dict[n]
else:
value = load_value(model, param_name, path)
set_module_tensor_to_device(model, param_name, device, value)
set_module_tensor_to_device(model, param_name, device, value, dtype=value.dtype)

return hook

Expand Down
Loading