Skip to content

b1294

Latest
Latest

Choose a tag to compare

View all tags
@akqmffl akqmffl released this 17 Jun 09:16
· 9 commits to main since this release
b1294
19bf9b8

Build Number: b1294
Operating System(s): windows
GPU Target(s): gfx1151
ROCm Version: bundled Windows ROCm runtime from the local llama.cpp MTP build
Llama.cpp Commit Hash: 267f8af
Build Date: 2026-06-17

This release provides a Windows ROCm gfx1151 package for AMD Ryzen AI Max Series
APUs with RDNA 3.5 graphics. It contains the custom llama.cpp MTP runtime,
Qwen3.6 MTP runtime settings, Lemonade integration presets, and launch/config
helpers.

Assets:

  • llama-b1294-windows-rocm-gfx1151-x64.zip: standalone llama.cpp runtime package.
  • lemonade-b1294-windows-rocm-gfx1151-x64.zip: Lemonade runtime overlay/full replacement package.

Direct download links:

Asset SHA-256:

llama-b1294-windows-rocm-gfx1151-x64.zip
7F441D546AD0EF0E3C00FA8FA8651EB15BA0C32D223E132A331721083C042C99

lemonade-b1294-windows-rocm-gfx1151-x64.zip
02D671D8E57C18A03C9608183FD09D5E806D22D3CFC37895340B5B335B21846D

Recorded benchmark anchors:

  • n=2 production lane: 128K context, 4096 generated tokens, FA off, 15.131 tok/s, 92.286% acceptance.
  • n=2 reasoning off: 128K context, 4096 generated tokens, FA off, 15.413 tok/s, 93.182% acceptance.
  • n=3 chat stream: 8192 context, 100 generated tokens, FA off, 18.050 tok/s, 97.333% acceptance, exact.

Primary server SHA-256:

7E70C443971F26DE6734444D7E459B9EF5136D9F96B480FDD4A3482190699284
Assets 4
Loading