当前位置：网站首页>Pruning - quantification - turn to onnx Chinese series tutorials

Pruning - quantification - turn to onnx Chinese series tutorials

2022-07-27 06:42:00 【LEILEI18A】

prune - quantitative - turn onnx Chinese series course

0.github Project links

Refer to ：https://github.com/gengyanlei/onnx2tensorRt

1. brief introduction

 notes ：
     author ：leilei
     This warehouse is used to record your own engineering practice ( prune 、 quantitative 、 turn onnx、tensorRt Import onnx-python), Support Chinese 
 notes ：
    tensorRt It's closed source ,deepstream、TensorRT-Inference-Server And so on are closed source ,TVM It's open source. ,
     Not yet in China 1 A unified , expect .
     Pruning is currently only used channel prune ！ Distillation is not considered for the time being ！

2. Environmental requirements

    darknet 
    mxnet1.6 + 
    pytorch1.5 + (yolov5 support pytorch1.4+)
    tensorRt7 + (tensorRt7 support python3.4-3.7)
    onnx1.5 + (tensorRt7 support onnx1.5)
    onnxruntime 1.0
    python3.6 +
    docker-cuda10.2 +
    ubuntu16.04 or ubuntu18.04

    TODO：trt8.0