Example how to retrain model on user data?
Could you please provide an example or dicumentation how to re-train existing model on user cifs?
Hi, thanks for your message.
Can you provide more details about what you are trying to do? If you would like to train a new model from scratch on your own set of CIFs, the instructions for preparing a custom CIF dataset are here.
However, fine-tuning an existing model on custom CIF files was not explored during development, and is outside the scope of this project. If you're interested in experimenting, I recommend checking out nanoGPT, which provides an example of fine-tuning a pretrained natural language model (e.g., on Shakespeare text). This approach might adaptable for CIFs.
Just to clarify, it is possible to fine-tune an existing model with the code in this repo. In your config file, set:
init_from: 'resume'
Make sure the out_dir points to the directory containing the pre-trained model checkpoint. Fine-tuning is typically done with a lower learning rate than was used during pre-training.
But, again, fine-tuning wasn't explored during the development of this project, so we can't say how well it will work in practice.