Cannot find checkpoint in eval_low_res_given_prompt.py

[zheminzhang96]

I tried to load the trained model weights obtained from your Google drive in trainer.load(-1) eval_low_res_given_prompt.py. But got the following errors:

[CugAgo]

Have you solved this problem？

[zheminzhang96]

Have you solved this problem？

yes.

[CugAgo]

I think in the load function, it is necessary to load the ckpt weight file, but the downloaded file is in pin format. Have you modified the code? How do you read the weights？
Thank you for your reply.

[zheminzhang96]

I think in the load function, it is necessary to load the ckpt weight file, but the downloaded file is in pin format. Have you modified the code? How do you read the weights？ Thank you for your reply.

Yes, I modified code. I provided the .bin path as checkpoint_path. Then used torch.load(...) instead of accelerator.load_state(...)

[CugAgo]

I think in the load function, it is necessary to load the ckpt weight file, but the downloaded file is in pin format. Have you modified the code? How do you read the weights？ Thank you for your reply.

Yes, I modified code. I provided the .bin path as checkpoint_path. Then used torch.load(...) instead of accelerator.load_state(...)

If use ema_model.load_state_dict() instead of accelerator.load_state(...). Does the accelerator in the model still need to load parameters? Do the accelerators in the model still need to load parameters, or do it also need to load the checkpoint？

Thank you！

[CugAgo]

I think in the load function, it is necessary to load the ckpt weight file, but the downloaded file is in pin format. Have you modified the code? How do you read the weights？ Thank you for your reply.

Yes, I modified code. I provided the .bin path as checkpoint_path. Then used torch.load(...) instead of accelerator.load_state(...)

If use ema_model.load_state_dict() instead of accelerator.load_state(...). Does the accelerator in the model still need to load parameters? Do the accelerators in the model still need to load parameters, or do it also need to load the checkpoint？

Thank you！

I have solved this problem and it seems that there is no need to load parameters for accelerator, so it can run correctly. Thank you