deepseek-v4.1IMAGE
DeepSeek V4.1 Flash EXL3 on dual DGX Spark: ~50 tok/s stable
Local deployment test shows the EXL3 quantized V4.1 Flash running stably around 50 tok/s on two DGX Spark units, with prompt cache working.
Taro L. Saito
@taroleoArchived 10 days agoOriginally posted 2026.10.18
Taro L. Saito @taroleoOriginally posted 2026.10.18
DeepSeek V4.1 Flash 1017 EXL3、2 x DGX Sparkでも安定して50 tok/sくらいは出て、prompt cacheも効いてるのでそこそこ使える
x.com/taroleo/status/2111704915760517351Archived 2026.10.18 16:31