Introduction
【免费下载链接】amct AMCT是CANN提供的昇腾AI处理器亲和的模型压缩工具仓。 
项目地址: https://gitcode.com/cann/amct
This project provides calling samples for different scenarios. After setting up the environment, you can try to run them according to actual scenarios:
| Sample | Algorithm | Description |
|---|---|---|
| Quantize model using MIN-MAX algorithm | Min-Max | Simple quantization based on extrema, best for beginners |
| Quantize model using AWQ algorithm | AWQ | Activation-aware weight quantization, suitable for large model PTQ |
| Quantize model using GPTQ algorithm | GPTQ | Weight quantization based on second-order information, layer-by-layer optimization |
| Quantize model using SmoothQuant algorithm | SmoothQuant | W8A8 quantization that smooths activation distribution |
| Quantize model using Cast direct conversion algorithm | Cast | HiFloat8 data direct conversion |
| Quantize model using Quantile algorithm | Quantile | HiFloat8 quantile quantization |
| Quantize model using OFMR algorithm | OFMR | Output feature Min-Max quantization |
| Quantize model using MXQuant algorithm | MXQuant | Micro-scaling floating-point quantization (MXFP8/MXFP4) |
| Quantize model using FlatQuant algorithm | FlatQuant | Quantization that flattens distribution through affine transformation |
【免费下载链接】amct AMCT是CANN提供的昇腾AI处理器亲和的模型压缩工具仓。 项目地址: https://gitcode.com/cann/amct
创作声明：本文部分内容由AI辅助生成（AIGC），仅供参考
