Blakey Wu

Thank you for your interest in our work! Which part of code are you wishing for?

HPC training code can be ready soon. I can also upload the LoRA training code, but the training data construction is relatively complex, and may take considerable time for me...

Hi Howard, I have just uploaded the LoRA training code. Please note that it hasn't been sanitized, and lacks necessary training environment specifications. I will provide the full training guide...

@VigneshBaskar I'm preparing the training script for the classifier. It will be ready in a few days.

I have updated the instructions for adapting SD, please check it out! I will also release the training code of the preference classifier in a few days, maybe also with...

Here it is: https://github.com/tgxs002/align_sd#training . @R2Bb1T

I train the LoRA model on 4 GPUs, fp16 precision. Other options are left as default. I haven't tried libraries like deepspeed, feel free to share your experience if you...

Working on that...

It takes a few hours to train 10k iterations with batch size 40. The exact time depends on the type of GPU.

Yes, the statistics are correct. You can try tuning the LoRA on HPD, but I feel the effect may be limited by the dataset size. The noise of the dataset...