Skip to content

Instantly share code, notes, and snippets.

View mmhorda's full-sized avatar

mmhorda

View GitHub Profile
@mmhorda
mmhorda / llama-cpp-qwen36-27b-q6-rx7900xtx.md
Last active August 2, 2026 09:42
Build and run llama.cpp for Qwen3.6 27B Q6 on RX 7900 XTX / ~25GB VRAM

llama.cpp Qwen3.6 27B Q6_K 131k Vulkan/RADV build and runbook

Exact disaster-recovery notes for rebuilding and starting the Qwen3.6 27B Q6_K llama.cpp server on an RX 7900 XTX-class AMD GPU.

This runbook documents the working 131k-context Vulkan/RADV setup:

  • llama.cpp commit: f449e0553708b895adbd94a301431cef691f632d
  • llama.cpp version: 9733 (f449e0553)
  • compiler: GNU/GCC 16.1.1
  • backend: Vulkan/RADV