tardellirs

tardellirs/gemma3moe-builder

Sparse-upcycle a dense Gemma-3-270m into a Mixture-of-Experts (gemma3moe) and serve it on llama.cpp (CPU). Upcycle, train, inspect routing, GGUF.

⭐ 43 ⑂ 0 Python NOASSERTION · 26 天前推送
43
Watchers
0
贡献者
0
Commits
0
Releases
0
Open Issues
26 天前
最近推送
原文 中文
暂无 README