deepseek-v4.1POST
本地跑 DeepSeek V4.1 1017：量化不同，模型体验也不同
用户指出单台 DGX Spark 本地跑的 DeepSeek V4.1 1017 因量化与配方差异，效果和订阅版并非同一个模型，本地对话质量难以比较。
Deseret Saint ۞⛩
@DeseretSaint10 天前收录原推发布于 2026.10.22
Deseret Saint ۞⛩ @DeseretSaint原推发布于 2026.10.22
Local AI model conversations are terrible to have because of how much different quants/recipes affect the quality of outcomes. When someone says they are running Deepseek V4.1 1017 locally on a single DGX Spark, that isn't the same Deepseek model your subscription covers.
这条是引用着另一条写的：x.com/italianclownz/status/2112986357505384722
x.com/deseretsaint/status/2113021433165250848采集于 2026.10.22 11:59