setting pretraining learning rate in command line interface

[2533245542]

I tried to set it by

litgpt pretrain --config pythia14m.yaml --optimizer.class_path torch.optim.AdamW --optimizer.init_args.lr 1e-5

but not working

[Kiwy3]

Hello,

To define optimizer parameters, the --optimizer argument takes dictionnary. I had the same problem, you can use for instance
-- optimizer '{'class_path': 'torch.optim.AdamW', 'init_args': {'lr': 0.001, 'weight_decay': 1e-05, 'betas': [0.9, 0.999]}} '