Baike.dev
Log in
Back to tool/Back to issues
#2341·llama-cpp-python

Support sm100 and sm120 in CUDA pre‑built wheels

Author: XuehaoSunCreated Jul 16, 2026Updated Jul 16, 2026
CUDA : ARCHS = 750,800,860,890,900

The current pre‑built CUDA wheels only support up to sm90. Please extend the build to include sm100 and sm120 so users with newer GPUs (B200 and RTX 50 series) can install directly via pip without compiling from source. Thank you!

Source: abetlen/llama-cpp-python

View original on GitHubView discussion on GitHub