search

microsoft / Cream

This is a collection of our NAS and Vision Transformer work.
MIT License
1.69k stars 230 forks source link

How to finetune EfficientViT #172

Closed xywlpo closed 1 year ago

xywlpo commented 1 year ago

hi, your work is amazing. If I want to finetune on my own datasets, how to set the training parameters, including epoch, lr and so on. Thanks for your help!

xinyuliu-jeffrey commented 1 year ago

Hi @xywlpo ,

Thanks for your interest. We finetune our model for 300 epochs, using AdamW optimizer, batchsize 256, learning rate 1×10−3 and weight-decay 1×10−8 for downstream classification datasets. The remaining hyperparameters are the same as the imagenet setting. Please try it.

Best, Xinyu .