/dalle2-laion

Pretrained Dalle2 from laion

Primary LanguagePython

πŸŽ¨πŸ€–πŸ¦
DALLE2 LAION

This is a collection of resources and tools for LAION's pre-trained DALLE-2 model. The official codebase can be found in DALLE2-PyTorch.

Awknowledgments

A big thanks to...

  • Zion & Aidan: For developing the training code in DALLE2-PyTorch, training the models, and writing the inference pipeline found here.
  • Romain Beaumont: For their project management skills and guidance throughout the project.
  • lucidrains: For spearheading the DALLE2 replication.
  • LAION: For providing support, feedback, and inspiration to the open-source AI community.
  • StabilityAI: For their generous donation of compute resources, without that these models would not exist.
  • Kumar: For fleshing out the initial training script for the DALLE2-prior.

How To Use

You can download the latest checkpoints, and use the inference pipeline available in this repository!

Pre-Trained Models

For access to the latest official checkpoints please visit the official πŸ€— huggingface repo.

If you are interested in viewing the training runs for these models, they can be found here decoder and prior. Please note that, due to the nature of rapid prototyping, these links may become outdated--we will try out best to keep them updated, however. If you notice that is the case, and would like up-to-date reports, feel free to open an issue!

Inference Scripts

Information on how to use the inference scripts can be found in the dalle2_laion folder README.md.

Community

If you make anything cool, we'd love to see! Join us on Discord Join us on Discord

Issues

If you have questions, run into specific problems, or have an idea that you think should be integrated to our inference pipeline, feel free to open an issue here on github.

Some things we ask you to please keep in mind πŸ˜„

  • Do not open issues ⚠️ in the DALLE2-PyTorch repo for any problems with code found here!
  • If you find a bug πŸ›, please report as much information as possible in your issue.
    • Things like your environment, library versions, and method of inference are imparitave to helping identify the issue.
  • If you have a feature you would like to add or request βœ‹, please give a brief and detailed description of its use case and why you believe it would be helpful.
  • Keep questions πŸ€” relevant to this repository, if you have more general questions about dalle2, diffusion, upscaling, or AI in general, please join our discord and find an appropriate channel!

DISCLAIMER:

DALLE2-LAION is not affilliated in any way with OpenAI. While the architecture closely follows the original paper, modifications have been made. The output of this model, opinions of the creators, and code provided does not reflect in any way that of OpenAI's.