Show HN: Fast NF4 dequantization Triton kernel (1.41x faster than bitsandbytes)

1 comments

• $DiabloD3 11 hours ago

  Neat, can't wait to see a llama.cpp PR for this.