search

AUTOMATIC1111 / stable-diffusion-webui

Stable Diffusion web UI
GNU Affero General Public License v3.0
138.33k stars 26.29k forks source link

[Feature Request]: Add Ascend NPU npu_fusion_attention to accelerate training #16199

Open kevin19891229 opened 1 month ago

kevin19891229 commented 1 month ago

Is there an existing issue for this?

What would your feature do ?

Proposed workflow

  1. Go to add description of npu_fusion_attention operator
  2. add Ascend NPU npu_fusion_attention to accelerate training

Additional information

No response

kevin19891229 commented 1 month ago

@AUTOMATIC1111

Coloured-glaze commented 4 weeks ago

Please use the following function to achieve the same effect


torch.nn.functional.scaled_dot_product_attention()