Current resume training does not record and continue the epoch and step number correctly

willisma / SiT

Official PyTorch Implementation of "SiT: Exploring Flow and Diffusion-based Generative Models with Scalable Interpolant Transformers"

https://scalable-interpolant.github.io/

MIT License

662 stars 35 forks source link

Current resume training does not record and continue the epoch and step number correctly #19

Closed xmhGit closed 2 months ago

xmhGit commented 3 months ago

Another question is about 'find_model' function and opt.load_state_dict(state_dict["opt"]), you may need to modify these two places when you are doing resume training.

willisma commented 2 months ago

Thanks for spotting this! We will update accordingly