q38rocm
Qwen 3.8 27B ROCmFP4 on AMD Strix Halo (Ryzen AI Max+ 395). Up to 36 tok/s via MTP Speculation, TurboQuant & Mesa RADV Wave64.
- amd-ryzen-ai
- fp4
- gguf
- llama-cpp
- mtp
- qwen
- radv
- rocm
- speculative-decoding
- strix-halo
- turboquant
- vulkan
- Stars
- 246
- Forks
- 12
- + today
- +1
- Created
- 2mo
Ranking data as of October 1, 2026 (UTC).
Overview
Qwen 3.8 27B ROCmFP4 on AMD Strix Halo (Ryzen AI Max+ 395). Up to 36 tok/s via MTP Speculation, TurboQuant & Mesa RADV Wave64. It ranks #5002 on GitTiger, gaining +1 star on October 1, 2026 (UTC).
The project is written in Python and has 12 forks. It was created 2mo ago.
Installation
git clone https://github.com/julianmb/q38rocm.git
cd q38rocm
# see README for setup