search

openai / guided-diffusion

MIT License
6.06k stars 807 forks source link

How to set hyper-parameters for training a diffusion model #46

Closed ewrfcas closed 2 years ago

ewrfcas commented 2 years ago

Should I set 'lr_anneal_steps' and 'weight_decay' or just keep them in default during training a diffusion model?

unixpickle commented 2 years ago

In our experiments, we did not use lr_anneal_steps or weight_decay.

Spidartist commented 6 months ago

@unixpickle so did you let it run and stop it whenever you want?

wwj723452017 commented 2 months ago

@Spidartist hi,did you figure out this problem?I've ran for two days and still no sign to stop,maybe I should set'lr_anneal_steps'?