deepseek-v4.1IMAGE

Project Maya v1.1.22 runs DeepSeek V4.1 Flash on 128GB RAM at ~1.6-bit

The experimental Maya-XS build keeps all 384 experts in 120.6GB of RAM, matches the original's next token 87.5% of the time with 8.7% higher loss, and needs about 280GB to download.

Colby

@dsoch8cArchived 8 days agoOriginally posted 2026.10.23

Project Maya v1.1.22 runs DeepSeek V4.1 Flash on 128GB RAM at ~1.6-bit

Colby @dsoch8cOriginally posted 2026.10.23

Open original on X

开源项目 Project Maya 成功把 DeepSeek-V4.1-Flash 完整跑在 128GB 内存的小主机上。北京时间10月22日晚发布的 v1.1.22 推出实验档 Maya-XS，将384个专家全部保留并做约1.6比特量化，全部常驻显卡可直接使用的内存。官方数据显示内存占120.6GB，161.3GB记忆表从固态硬盘读。下载量发布正文写276.6GB、模型卡写280.9GB；作者自测相同下一个词占87.5%，下一词损失比原模型高8.7%。该档位在发布时仍标为实验性，87.5%也并非标准跑分，且完整下载需要约280GB。 https://github.com/mw00/project-maya/releases/tag/v1.1.22

x.com/dsoch8c/status/2113603409886752848Archived 2026.10.24 12:08