Skip to content

GPTFast 0.3.0

Choose a tag to compare

@MDK8888 MDK8888 released this 21 Jun 01:51
60ab0b8
  • GPTQ INT4 quantization available for all HF models
  • Accelerates inference speed by 7.6x-9x
  • Integrates optimized INT4 matrix multiplication kernels from the PyTorch team for all HF models