AI Resonance · 阅读最新日报 · AI 入门推荐

2026-11-15 · America/Los_Angeles · 社区动态 · #19

BigMoeOnEdge outperforms Strata and colibri on 8-16 GB cards and Mac and you can run any model

Hi! for those who might not know the solution: it takes any MoE GGUF llama.cpp can load, whatever the quant. No repacking, no special format it runs on whatever you have: NVIDIA, AMD, Intel, Mac, a phone, or just the CPU Windows, Linux, macOS, Android, iOS output is the same as llama.cpp, don't cut corners on quality you start it with --auto and it works out the settings for your machine Below is a brief overview of the benchmark: 8 GB card, Qwen3.6-35B: I get about 35 tok/s. colibri around 23. Strata wouldn't start at 16k context 12 GB card, same model: about 45 for bmoe, about 30 for…

BigMoeOnEdge outperforms Strata and colibri on 8-16 GB cards and Mac and you can run any model

热度 42.6 / 100;排名与评分保留该期记录。