163 6 days ago

100% Uncensored (9/9 SecOps & NSFW, MIT License) Moonshot AI Kimi-Linear 48B-A3B Hybrid KDA Linear-Attention MoE (3B active, 1M token ctx, BF16-Abliterated Q4_K_M) — 142.3 tok/s decode in 28.1 GiB VRAM on A100 40GB.

48b