I'm trying to train the model using this command:
torchrun --standalone --nproc_per_node 1 /path/opensora/Open-Sora/scripts/train.py /path/opensora/my_train_config.py .
but i'm getting this error even though my csv file is in this format : path,id,relpath,num_frames,height,width,aspect_ratio,fps,resolution,text
Traceback (most recent call last):
File "/home/txt_vid/opensora/Open-Sora/scripts/train.py", line 423, in <module>
main()
File "/home/txt_vid/opensora/Open-Sora/scripts/train.py", line 315, in main
loss_dict = scheduler.training_losses(model, x, model_args, mask=mask)
File "/home/txt_vid/opensora/Open-Sora/opensora/schedulers/rf/__init__.py", line 103, in training_losses
return self.scheduler.training_losses(model, x_start, model_kwargs, noise, mask, weights, t)
File "/home/txt_vid/opensora/Open-Sora/opensora/schedulers/rf/rectified_flow.py", line 89, in training_losses
t = timestep_transform(t, model_kwargs, scale=self.transform_scale, num_timesteps=self.num_timesteps)
File "/home/txt_vid/opensora/Open-Sora/opensora/schedulers/rf/rectified_flow.py", line 23, in timestep_transform
if model_kwargs[key].dtype == torch.float16:
KeyError: 'height'
Hello guys,
I'm trying to train the model using this command:
torchrun --standalone --nproc_per_node 1 /path/opensora/Open-Sora/scripts/train.py /path/opensora/my_train_config.py .
but i'm getting this error even though my csv file is in this format :
path,id,relpath,num_frames,height,width,aspect_ratio,fps,resolution,text
and this is my config :