Skip to content

llama.cpp ROCm v0.0.4-experimental [Experimental RDNA Boosts (Universal AVX2)]

Pre-release
Pre-release

Choose a tag to compare

View all tags
@github-actions github-actions released this 10 Oct 23:20
· 30 commits to custom-workflow since this release
v0.0.4-experimental
1ebeee7

Release Information

  • Flavor: Experimental RDNA Boosts (Universal AVX2)
  • Source Branch: experiment/rdna-boosts
  • Commit: ce7417cf78cb0e198c38f3acdeaf4dadd19ce9cc
  • ROCm Core SDK: 10.1.0 (TheRock)
  • Target GPU Architectures: gfx900;gfx906;gfx908;gfx90a;gfx1030;gfx1100;gfx1101;gfx1102;gfx1200;gfx1201
  • CPU Target: Standard AVX2 (Universal x86-64 compatibility)

Description

TurboQuant KV cache + Stew's RDNA Boosts (native-BF16 FA, WMMA, fused GDN, fused MoE) + ROCmFPX quantization family + DFlash2 speculative decoding. Universal AVX2 build compatible with all modern x86-64 CPUs.

Unified ROCm 10.1.0 (TheRock) release built for Windows and Linux.
Packages include required AMD ROCm runtime libraries (rocblas, hipblaslt, hipblas, amdhip64) for portable out-of-the-box execution.

  • Windows Scripts: llama-rocm-experimental-scripts-windows.zip (PowerShell quantization scripts for Windows)
  • Linux Scripts: llama-rocm-experimental-scripts-linux.zip (Bash quantization scripts for Linux)
Assets 6
Loading