sogh5/SubsetQ

Implementation for Non-Uniform Step Size Quantization for Accurate Post-Training Quantization (ECCV 2022)

PythonMIT

Subset Quantization

Implementation for Non-Uniform Step Size Quantization for Accurate Post-Training Quantization (ECCV 2022)

Requirements

Python v3.6
PyTorch 1.7.1
CUDA 11.1 with cuDNN 8 (We use NVIDIA GeForce RTX 3090 GPU)

How to Run Subset Quantization

./run_ptq.sh [DATASET_PATH] [WBIT] [GPU_ID]

How to Run Evaluation

./test_ptq.sh [DATASET_PATH] [MODEL_FILE_PATH] [GPU_ID]

etc.

Currently, we support only single GPU mode.
Our codes are based on implementation: https://github.com/yhhhli/BRECQ (Copyright (c) 2021 Yuhang Li, under MIT License).
- Please refer to the files: NOTICE.md, BRECQ-LICENSE