Training settings #13
Description
During training phases one and two, how do I select ./configs/model/stage1/xx.yaml and ./configs/model/stage2/xx.yaml to ensure that the trained model matches the five data points you provided?
Also, do I need to use checkpoints in the resume during phase 1?Because the running results indicate [329 keys are missing from the model (hint processing and cross connections included)]
How to use the model trained in stage 1 when trying to train stage 2？
Since there is no testing process, how can I determine if training is complete and what the optimal weights are?
I noticed that the parameter settings in the training code differ somewhat from those mentioned in the paper. Should I use the code directly?
Is the missing setting in the PFO test code correct?
clip_model, _, clip_preprocess = open_clip.create_model_and_transforms("ViT-H-14", pretrained="weight/open_clip_pytorch_model.bin", device=model.device)
Additionally, the failure to initialize the opt parameter causes errors in prompt_inversion/test_optim_zc.py, such as opt.iter, opt.clip_model, etc.
@KangLEkk Thank you for taking the time to answer my questions.