* is some error
```
[rank1]: File "/mnt/ceph/develop/jiawei/cogvideox-distillation/training/cogvideox_image_to_video_lora.py", line 1005, in <module> [rank1]: main(args) [rank1]: File "/mnt/ceph/develop/jiawei/cogvideox-distillation/training/cogvideox_image_to_video_lora.py", line 884, in main [rank1]: "gradient_norm_before_clip": gradient_norm_before_clip, [rank1]: UnboundLocalError: local variable 'gradient_norm_before_clip' referenced before assignment
```
ref1 :https://github.com/a-r-r-o-w/cogvideox-factory/pull/84
ref2: https://github.com/a-r-r-o-w/cogvideox-factory/pull/100
It seems that there is a bug in the accelerator.is_main_process task and accelerator.distributed_type. Is_main_process has scheduling problems during the training initiation and training stages.
* fix
* adaption for CogVideoX1.5
* add patch_size_t in full finetuning of T2V and lora finetuning of I2V
* Update training/args.py
Co-authored-by: Sayak Paul <spsayakpaul@gmail.com>
---------
Co-authored-by: Sayak Paul <spsayakpaul@gmail.com>
* replaced lambda statements with local functions to work with pickle in dataset.py and moved collate function outside of main and added its own class in cogvideox_image_to_video_lora.py
* fixed merge issue
* fixed row deletion merge
* fixed row deletion merge
* fixed comma-separated param issue in bash
* fix: resuming from a checkpoint when using deepspeed.
* remove changes to prepare_dataset.py
* propagate to others.
* tackle gradnorm.
---------
Co-authored-by: Aryan <aryan@huggingface.co>
glide-the