Error when using batch_size.

[neph1]

System Info / 系統信息

Thank you for providing a way to finetune LTX-video. Training seems to work well, but when I try changing batching (in this case to 4), I encounter this exception. I'm using the suggested example script in the repo. Running on diffusers from github, updated two days ago.

This is the exception:

12/21/2024 12:38:27 - ERROR - finetrainers - Traceback (most recent call last):
  File "/some_path/train.py", line 34, in main
    trainer.train()
  File "/some_path/trainer.py", line 424, in train
    noisy_latents = (1.0 - sigmas) * latent_conditions["latents"] + sigmas * noise
RuntimeError: The size of tensor a (4) must match the size of tensor b (128) at non-singleton dimension 2

Also hijacking my own issue:
I tried adding --resume_from_checkpoint since that worked with the old cogvideox-factory trainer. But it doesn't seem to do anything now. Is that correct?

Thanks!

Information / 问题信息

• The official example scripts / 官方的示例脚本
• My own modified scripts / 我自己修改的脚本和任务

Reproduction / 复现过程

Change --batch_size 1 to something other than 1

Expected behavior / 期待表现

Training would continue with batching enabled

[a-r-r-o-w]

Oh interesting, thanks for reporting! I'll look into this shortly.

--resume_from_checkpoint support will be added soon too. I think #130 has it working, so it will be in main soon hopefully