Skip to content
/ ComfyUI-WanVideoWrapper-ultravico_AMD Public

Modified sageattn_ultravico for AMD/ROCm compatibility with `triton-windows`

0 stars 0 forks Branches Tags Activity
Star
Notifications You must be signed in to change notification settings

0xDELUXA/ComfyUI-WanVideoWrapper-ultravico_AMD

BranchesTags

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

22 Commits
README.md
README.md
 
 
attn_qk_int8_per_block.py
attn_qk_int8_per_block.py
 
 
core.py
core.py
 
 
quant_per_block.py
quant_per_block.py
 
 

Repository files navigation

View Changes

You can see the exact modifications made for AMD and triton-windows compatibility here.

Technical Changes:

  • Block sizes modified for better ROCm compatibility:
    • BLKQ: 128, BLKK: 64
    • BLOCK_M: 128, BLOCK_N: 64
  • Numerical Stability: Kept accumulation in float32.
  • Optimization: Triton autotuning added.

Tested on RDNA 4 with TheRock ROCm 7, PyTorch and triton-windows release v3.6.0-windows.post25.

Installation (PowerShell)

cd ComfyUI/custom_nodes/ComfyUI-WanVideoWrapper/ultravico/
Remove-Item -Recurse -Force sageattn
git clone https://github.com/0xDELUXA/ComfyUI-WanVideoWrapper-ultravico_AMD sageattn

About

Modified sageattn_ultravico for AMD/ROCm compatibility with `triton-windows`

Resources

Readme
Activity

Stars

0 stars

Watchers

0 watching

Forks

0 forks
Report repository

Releases

No releases published

Packages

 
 
 

Contributors

Languages