Skip to content

fix xpu ipex linear in torch2.7 #1618

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
Titus-von-Koeller merged 1 commit into from
Apr 29, 2025
Merged

fix xpu ipex linear in torch2.7 #1618

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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: 7 additions & 1 deletion bitsandbytes/utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -240,10 +240,16 @@ def enable_ipex_fusion(linear, x):
)
elif x.device.type == "xpu" and ipex_xpu and _ipex_xpu_version_prereq(2, 5):
converted_weight = reverse_4bit_compress_format(linear.weight.data)
new_weight = converted_weight.reshape([quant_state.shape[0], quant_state.shape[1] // 2])
new_scales = quant_state.absmax.view(quant_state.shape[0], quant_state.shape[1] // quant_state.blocksize)
new_zeros = None
compensation = None
new_weight = converted_weight.reshape([quant_state.shape[0], quant_state.shape[1] // 2])
# ipex 2.7 requires new_scales is a list of tensors
if _ipex_xpu_version_prereq(2, 7):
new_scales = list(new_scales)
# ipex 2.7 can dequant converted_weight directly.
if linear.training or x.requires_grad == False:
new_weight = converted_weight
else:
raise ValueError(
"Please check the device and ipex version. The device should be cpu or xpu while ipex version should >= 2.5"
Expand Down