After training diffusion model with human preference dataset.

[giovanlee]

Thanks for your research.

I have question with training d3po with my diffusion model.

I trained my diffusion model with scripts/train.py, and I see output files are at logs/checkpoint_0. There are optimizer.bin, pytorch_lora_weights.bin, random_states_0.pkl, scaler.pt files. After training model, how can I sample images from trained model?

Regrads,

[yk7333]

To load the training model, you can modify config.resume_from in the config file to point to your output files.

[hdeng26]

I opened a pr for fixing the model load checkpoint from local dir but only pytorch_lora_weights.bin saved in training script

[giovanlee]

Thanks for reply. I used ur fixed scripts/sample.py code to sample images from my local checkpoint files, such as logs/run_name/checkpoints/checkpoint_165 in config.resume_from. However, I have an error with AttributeError: 'AttnProcsLayers' object has no attribute 'load_attn_procs'. Do you have any idea about this error?

Regards,