search

huqinghao / PalQuant

12 stars 1 forks source link

default config training failed #1

Closed ChuanjunLAN closed 2 years ago

ChuanjunLAN commented 2 years ago

Thx for your work, i tried to reimplement your repo by default config you mentioned : "python main.py --data your-imagenet-data-path --visible_gpus '0,1,2,3' --workers 20 --arch 'resnet18_quant' --epochs 90 --groups 2 --weight_levels 4 --lr_m 0.1 --lr_q 0.0001 -b 256 --act_levels 4 --log_dir "../results/resnet-18/W2A2G2/" but i cant get a even close result as you put in your paper that quantization level is 2 bits for Resnet18 imagenet dataset, i think your config above is for 4 bits. so is there some setup i missed? wait for your help, thx again

huqinghao commented 2 years ago

After refactoring the codes, I only evaluate the model but haven't re-train the model. I will check the codes and args. BTW, could you post the printed args or the log.txt in the log-dir?

huqinghao commented 2 years ago

I have made some updates, and the training logs and models are also uploaded. Now you can re-run the code with the default config.

ChuanjunLAN commented 2 years ago

I have made some updates, and the training logs and models are also uploaded. Now you can re-run the code with the default config.

Thx for you reply, it works well for the default config. Have a good day