今天新增deepseek-v4.1POST
DeepSeek V4.1 Flash 原生量化在 4 台 DGX Spark 上跑出可用速度
用户反驳「小模型 4-bit 量化才跑得动」的说法，指出 V4.1 Flash 1017 的完整原生量化可在 4 台 DGX Spark 上以不错的速度运行。
Random Libertarian Tech Lead
@someRandomDev58 小时前收录原推发布于 2026.10.28
Random Libertarian Tech Lead @someRandomDev5原推发布于 2026.10.28
@OliverDlouhy @log4j @huggingface Somehow I think you've only tried out e.g. 4-bit quants of small models that can fit on 16-32GB MacBooks, rather than something like "the full native quant of DeepSeek V4.1 Flash 1017, which can run at solid speeds across 4x DGX Sparks".
x.com/somerandomdev5/status/2115156516311605255采集于 2026.10.28 12:04