⬢github Python · 48 ★ +48 since we first saw it · pushed 2 h ago · NOASSERTION
AmSach/maple
MAPLE: Manifold-Activated Pruning of Latent Experts for Sub-2-BPW MoE LLMs
MAPLE is a Python pruning pipeline for Mixture-of-Experts LLMs that profiles which experts are rarely activated for a given domain, removes dormant ones, and packs the compressed model into a low-bit GGUF. The released checkpoint compresses Gemma 4 26B-A4B to 4.75 GB (sub-2 bits/param), so it fits fully in a 6 GB GPU's VRAM and runs via llama.cpp.
Why now: It was just shared on Hacker News as a Show HN demo, with a headline claim of running a 26B MoE model at 45 tokens/second on a 6GB consumer GPU.
Who it is for: Hobbyists and developers who want to run large MoE models for coding/technical tasks on low-VRAM consumer GPUs.
Stars over our 398 snapshots: 0 to 48, since 4 d ago.
Where people talked about it
API: https://socialmediatrends-api.osmike.com/v1/repos/AmSach/maple