Skip to content

Latest commit

 

History

History
8 lines (6 loc) · 321 Bytes

README.md

File metadata and controls

8 lines (6 loc) · 321 Bytes

ResMLP_QAT

Quantization-aware training research for ResMLP model.

Finding initial scale for v2

python quant_train.py -a q_resmlp_v2 --epochs 3 --lr 0.001 --batch-size 16 --data /mnt/disk1/imagenet/ --pretrained --save-path folder/checkpoints/ --wd 1e-4 --data-percentage 0.002 --gpu 1 --freeze-w --wandb