Repository navigation
b1381
| gfx family | Windows | Linux |
|---|---|---|
| gfx1151 | zip · 196 MB | zip · 659 MB |
| gfx1152 | zip · 191 MB | zip · 635 MB |
| gfx1153 | zip · 191 MB | zip · 635 MB |
| gfx1150 | zip · 193 MB | zip · 637 MB |
| gfx117X gfx1170 · gfx1171 |
zip · 472 MB | zip · 935 MB |
| gfx120X gfx1200 · gfx1201 |
zip · 695 MB | zip · 1.19 GB |
| gfx110X gfx1100 · gfx1101 · gfx1102 · gfx1103 |
zip · 540 MB | zip · 1.11 GB |
| gfx103X gfx1030 · gfx1031 · gfx1032 · gfx1034 |
zip · 547 MB | zip · 1.15 GB |
| gfx90a | zip · 338 MB | zip · 799 MB |
| gfx908 | zip · 226 MB | zip · 687 MB |
- Every archive passed the release checks.
- Each archive bundles its ROCm runtime, so nothing else needs to be installed.
- If your GPU's gfx target isn't listed here, these builds don't support it.
Build Number: b1381
ROCm Version: 10.3.0a20261118
Llama.cpp Commit Hash: f8caa
Llama.cpp Commit (full): f8caa5ea30b5775512906ae3b7ee53b21a7a6e1d (view on GitHub)
Build Date: 2026-11-19 14:08:47 UTC
Verify your download
checksums.txt holds the SHA-256 of every archive.
- Linux:
sha256sum --ignore-missing -c checksums.txtin the folder holding the archive andchecksums.txt. - Windows (PowerShell):
Get-FileHash <archive>.zip -Algorithm SHA256, then compare the hash with the archive's line inchecksums.txt(case does not matter).
qualification-summary.json records why this release was published. Each archive also has a qualification-<archive>.json in the assets below, listing what it contains, which gfx target (if any) it was run on real hardware with, and every check's result.