sndnyang / DiffusionRecoveryLikelihood-ipynb

Replace the global absl.flags variable to the local variable hps, then create the ipynb for jupyter/colab

Learning Energy-Based Models by Diffusion Recovery Likelihood

I don't want the global variable FLAGS from absl.flag.

I create a ipynb notebook difrec_tf2.ipynb and verify it on colab. Today is 2023.01.27

You need to upload the related *.py files to colab, and then

you can change the settings and note that

1. the default model T6 with (6,30) is slow.
2. (1000, 0) will fail, so I don't know how to train T1k

hps['num_res_blocks'] = 2
hps['num_diffusion_timesteps'] = 6
hps['mcmc_num_steps'] = 30

Just click the button or go to check the file (a colab button in it)

One more thing

https://github.com/sndnyang/DiffusionRecoveryLikelihood-PyTorch

The pytorch implementation (I hope it can run smoothly but there may exists bugs)

one more thing

Will upload T1k setting soon! --- 2 years later

Official Repo

https://github.com/ruiqigao/recovery_likelihood

If you find their work helpful to your research, please cite:

@article{gao2020learning,
  title={Learning Energy-Based Models by Diffusion Recovery Likelihood},
  author={Gao, Ruiqi and Song, Yang and Poole, Ben and Wu, Ying Nian and Kingma, Diederik P},
  journal={arXiv preprint arXiv:2012.08125},
  year={2020}
}