Skip to content

[XPU] support bf16 weight_scale in op weight_only_linear #73963

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 1 commit into from
Jul 16, 2025
Merged

[XPU] support bf16 weight_scale in op weight_only_linear #73963

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
22 changes: 21 additions & 1 deletion paddle/phi/kernels/xpu/weight_only_linear_kernel.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -60,14 +60,34 @@ void WeightOnlyLinearKernel(const Context& dev_ctx,
input_y, nullptr, m, n, n, false};
baidu::xpu::xblas::FcFusionTensor<XPUType> tensor_y{
input_y, nullptr, m, n, n, false};
DenseTensor weight_scale_fp32;
if (weight_scale.dtype() != phi::DataType::FLOAT32 &&
weight_scale.dims().size() != 0) {
weight_scale_fp32.Resize(weight_scale.dims());
dev_ctx.template Alloc<float>(&weight_scale_fp32);
int r = baidu::xpu::api::cast<XPUType, float>(
dev_ctx.x_context(),
reinterpret_cast<const XPUType*>(weight_scale.data<T>()),
weight_scale_fp32.data<float>(),
weight_scale.numel());
PADDLE_ENFORCE_XDNN_SUCCESS(r, "cast");
}
const float* weight_scale_ptr = nullptr;
if (weight_scale.dims().size() != 0) {
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

为什么上面不需要考虑weight_scale.dims().size() != 0？这种情况调用cast会出错吗？
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

考虑scalar tensor，dims().size()可能为0，但是numel为1，考虑是否是从numel来判断
Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

原有实现中dims().size()等于零时直接传nullptr进算子，所以不需要考虑scalar。cast出错风险我修一下
if (weight_scale.dtype() == phi::DataType::FLOAT32) {
weight_scale_ptr = weight_scale.data<float>();
} else {
weight_scale_ptr = weight_scale_fp32.data<float>();
}
}
baidu::xpu::xblas::FcFusionEpilogue<float, float> epilogue{
api::Activation_t::LINEAR,
bias.is_initialized() ? (bias.get().dtype() == phi::DataType::FLOAT16
? bias_fp32.data<float>()
: bias.get().data<float>())
: nullptr,
nullptr,
weight_scale.dims().size() != 0 ? weight_scale.data<float>() : nullptr,
weight_scale_ptr,
0,
1,
nullptr};
Expand Down
Loading