sci-ml/saber
Rust LLM inference engine for AMD Radeon AI PRO R9700 (gfx1201 / RDNA4)
-
saber-0.0.1~amd64debug
View
Download
Browse License: Apache-2.0 BSD ISC MIT ZLIB Unicode-3.0 || ( Apache-2.0 BSD-2 ) || ( Apache-2.0 MIT ) || ( BSD-2 MIT ) Overlay: bennypowers
ChangeLog
commit a2835d15c2afc022b854fe30ab6c44c17298a5a9
Author: Benny Powers <web@bennypowers.com>
Date: Sat Nov 7 21:19:05 2026 +0200
feat(sci-ml/saber): add saber 0.0.1
Rust LLM inference engine for AMD Radeon AI PRO R9700 (gfx1201 /
RDNA4), from GGZ14/saber-engine. No release tags upstream, so the
ebuild pins a commit and .larry/prepare.sh vendors the workspace
crates into a tarball uploaded to the package-named release.
Kernels are prebuilt HIP code objects installed to
/usr/share/saber/kernels/gfx1201 (pass --prebuilt to saber); the
HIP runtime is dlopen'd at run time via --hip-lib, so there is no
ROCm build or runtime dependency.
Author: Benny Powers <web@bennypowers.com>
Date: Sat Nov 7 21:19:05 2026 +0200
feat(sci-ml/saber): add saber 0.0.1
Rust LLM inference engine for AMD Radeon AI PRO R9700 (gfx1201 /
RDNA4), from GGZ14/saber-engine. No release tags upstream, so the
ebuild pins a commit and .larry/prepare.sh vendors the workspace
crates into a tarball uploaded to the package-named release.
Kernels are prebuilt HIP code objects installed to
/usr/share/saber/kernels/gfx1201 (pass --prebuilt to saber); the
HIP runtime is dlopen'd at run time via --hip-lib, so there is no
ROCm build or runtime dependency.

