bitsandbytes Files

Highlights

• ROCm 10.0 build is now included for both Linux and Windows.
• Additional AMD GPU targets are included in our prebuilt wheels.
• Performance improvement for certain 4bit GEMM shapes on NVIDIA GB10.

What's Changed

• Mask the absmax store in the Triton 4-bit quantize kernels by @truong-v in https://github.com/bitsandbytes-foundation/bitsandbytes/pull/2043
• Add sm121 (GB10) to the wide_n_simt 4-bit GEMM dispatch rule by @yashb98 in https://github.com/bitsandbytes-foundation/bitsandbytes/pull/2039
• chore(deps): bump the actions group across 1 directory with 2 updates by @dependabot[bot] in https://github.com/bitsandbytes-foundation/bitsandbytes/pull/2036
• [ROCm] Harden HIP version for native library naming and selection by @sstamenk in https://github.com/bitsandbytes-foundation/bitsandbytes/pull/1888
• Revert "[ROCm] Harden HIP version for native library naming and selection" by @matthewdouglas in https://github.com/bitsandbytes-foundation/bitsandbytes/pull/2060
• [ROCm] Restore Wave64 warp size for all gfx9 targets by @0xDELUXA in https://github.com/bitsandbytes-foundation/bitsandbytes/pull/2059
• ci: expand ROCm architecture coverage by @sstamenk in https://github.com/bitsandbytes-foundation/bitsandbytes/pull/2046
• [ROCm] Add CI for ROCm 10.0 by @sstamenk in https://github.com/bitsandbytes-foundation/bitsandbytes/pull/2061

New Contributors

• @truong-v made their first contribution in https://github.com/bitsandbytes-foundation/bitsandbytes/pull/2043
• @0xDELUXA made their first contribution in https://github.com/bitsandbytes-foundation/bitsandbytes/pull/2059

Full Changelog: https://github.com/bitsandbytes-foundation/bitsandbytes/compare/0.50.1...0.50.2