deepseek-v4.1POST
4 台 DGX Spark 本地跑 V4.1 Flash：97 tok/s、29.4M KV 池
用 TensorFold v0.3.5 配方与 4.0bpw 量化在 4 台 DGX Spark 部署，单用户 97 tok/s、16 用户 231 tok/s，93K 提示读取 37 秒。
Tech2Wild
@Tech2Wild1 天前收录原推发布于 2026.10.27
Tech2Wild @Tech2Wild原推发布于 2026.10.27
DeepSeek V4.1 Flash on 4x DGX Spark, now on @bertholomusai's TensorFold v0.3.5 recipe + his 4.0bpw pack: • 29.4M-token KV pool (was 2M) • 97 tok/s at 1 user, 144 on code, 231 at 16 users • 93K prompt read in 37s, ~2s replies during long reads https://github.com/tonyd2wild/DeepSeek-V4.1-Flash-TP2-and-TP4-DGX-Sparks
x.com/tech2wild/status/2114866340645278008采集于 2026.10.27 12:08