Show HN: Fast NF4 dequantization Triton kernel (1.41x faster than bitsandbytes)(github.com)
3 pointsby Griffith-78 hours ago1 comment
  • DiabloD32 hours ago
    Neat, can't wait to see a llama.cpp PR for this.