train.py

[hhb1224]

Hi, Thank you for approaching me efficient pruning method.

I implemented 1 of 2 pruning steps(Training & Masked retraining). While I train with train.py, 4 batch size, and 25 epoch, It implements 0-24 steps and again and again.. When will it stop itself? and what does this iteration means?

[nightsnack]

Hi hhb1224,
The first step includes 4 iterations of rho (in cfg/darknet_admm.yaml: --rou-num) each iteration uses a different rho value. Starting from 0.0001, the rou value increases by 10x each iteration, which means the punishment of pruning loss increases by 10x each iteration. For more details of admm algorithms please check this paper: A Systematic DNN Weight Pruning Framework using Alternating Direction Method of Multipliers https://arxiv.org/pdf/1804.03294.pdf

[baisong666]

我第一步训练完后生成的4个pt文件都是256M，这正常吗，第二步最后生成的模型文件竟然有500M那末大，请问为什幺呢，谢谢拉