Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

Its about 20-40% depending on the GPU, from my tests.

And only very recent builds of torch 2.1 (with dynamic input) work properly, and it still doesn't like certain input changes, or augmentations like controlnet.

AIT is the most usable compiled implementation I have personally tested, but SHARK (running IREE/MLIR/Vulkan) and torch-mlir are said to be very good.

Hidet is promising but doesn't really work yet. TVM doesn't have a complete implementation outside of the WebGPU demo.



Try head of master. If there’s any bugs or graph breaks you hit, lmk, I can take a look. My numbers say 71% with a few custom hacks.

Glad the dynamic stuff is working out tho!


I will, thanks!

I have been away for a month, but I will start testing it again later and submit some issues I run into.


My username without underscore, at meta. Email me any bugs, I can help file them on GH and lend a hand fixing.




Consider applying for YC's Fall 2026 batch! Applications are open till July 27.

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: