sleepyeldrazi/ds4-nvfp4-spark

Mixed NVFP4 serving of DeepSeek V4 Flash on DGX Spark (GB10) - fork of antirez/ds4 with REAP expert pruning, NVFP4 quantization, FP8-packed KV cache, and managed-memory serving

3 best points · 1 comments · 1 Hacker News post · 8 GitHub stars · View on GitHub

Hacker News discussions

• DeepSeek V4 Flash optimized framework and model variants for DGX Spark (3 points, 1 comments, by sleepyeldrazi, 2026-06-22)