Tools
Edge0 runs a 35B MoE on Apple Silicon in 2.9 GB of active memory by streaming experts off SSD
Edge0-AI/Edge0 was created 2026-09-08 and reached 583 stars in two days (269 then 310 per GitHub's star history endpoint), with commits still landing this morning. It packages SSD expert offload, Recover-LoRA, and prerouter routing prediction into an MLX-backed framework: `edge0-35b` is a 4-bit 40-layer 256-expert model built on Qwen3.5-MoE 35B-A3B needing ~2.9 GB peak active memory against a ~23 GB on-disk checkpoint, scoring 79.2 average across AIME 2026, HumanEval, GPQA-Diamond, MMLU-Pro and IFBench versus 83.2 for the fp16 base. A CUDA backend is a reserved directory, not yet shipped.
Source
↳ Follow the thread