freriadoc / Avalon
git clone https://github.com/freriadoc/Avalon.git
Fast llama.cpp fork that runs the Qwen3.8-Flash-Next 125B MoE model on one consumer AMD GPU by streaming experts from system RAM.
Why you might care
Targets a single model with a GPU-plus-RAM expert cache, reporting 91 tokens/s on an RX 7900 XT with 32 GB RAM enough to run an 89 GiB model. Useful if you want to run this specific large MoE on ROCm hardware; benchmarks are self-reported and it is not a general-purpose llama.cpp replacement.
+66
stars, 24h · #662
+~581
stars, 7 days · #473
+193
stars, 30 days · #5,324
193
★ total · +581% this week
32
forks · 13 open issues
Star history (as recorded here)
LanguageC++
LicenseMIT
Created2026-10-09 (23d ago)
Last pushtoday
Watchers4
OwnerUser
Links
README
Loading README…
Blurb generated by claude-haiku-5-5 today. Star deltas marked ~ are extrapolated from partial history.