Dual Spark Runs Abliterated DeepSeek Locally at 40 tok/s

Developer Teknium connected two NVIDIA DGX Spark devices with a single cable to run an abliterated DeepSeek V4 Flash 0731 model locally, achieving 40 tokens per second without dspark, enabling fully uncensored inference.

2026-08-09 ~ 2026-08-09 · 3 related posts

• Dual DGX Spark Setup Hits 40 tok/s on DeepSeek Locally — Teknium · 2026-08-09
• Running DeepSeek Locally on Dual Sparks: 40 tok/s Uncensored Inference — Teknium · 2026-08-09
• Dev Runs Uncensored DeepSeek Locally Using Dual Sparks at 40 tok/s — PMinervini · 2026-08-09