deepseek-v4.1IMAGE

Uncensored DeepSeek V4.1 Flash at 3.5bpw hits 50 tok/s on 7x3090

An EXL3 3.5bpw uncensored quant of DeepSeek-V4.1-Flash-1017 runs at about 50 tok/s decode on 7x3090 plus 256GB RAM using the Project Maya engine, with roughly 260GB of weights.

Axitek

@axitekArchived 10 days agoOriginally posted 2026.10.22

Uncensored DeepSeek V4.1 Flash at 3.5bpw hits 50 tok/s on 7x3090

Axitek @axitekOriginally posted 2026.10.22

Open original on X

DeekSeek-v4.1-Flash-1017-Uncensored at 3.5bpw, running on Maya v1.1.20 at 50 tps decode 🚀 This is HUGE - like, 260gb weights + 200gb engrams, running on 168gb vram - all thanks to Project Maya. Model: https://huggingface.co/neko-legends/DeepSeek-V4.1-Flash-1017-EXL3-3.5bpw-UNCENSORED-4xSpark Engine: https://github.com/mw00/project-maya H/W 7x3090+256gb

x.com/axitek/status/2112996349000192267Archived 2026.10.22 12:01