wont start train

[theaorora]

Hi tdrussell,
I followed the instruction but got this error

[2024-12-23 05:32:43,472] [INFO] [real_accelerator.py:222:get_accelerator] Setting ds_accelerator to cuda (auto detect)
[2024-12-23 05:32:47,733] [INFO] [launch.py:351:main] Process 3676 exits successfully.
(diffusion-pipe) root@5c5c3dc5ac6c:/workspace/diffusion-pipe#

my datasets is only images without caption.txt is this related to the problem? (i train person with flux before without .txt and no problem at all)
if i want to use video in datasets what format should i put? mp4?
can i mix images and videos in datasets?

[tdrussell]

I haven't seen this before. It looks like something is going wrong with deepspeed, it doesn't even launch the script, just immediately exits. What is the deepspeed command you're using to run it?

For video, I have only tested with mp4 but anything that the imageio library can load should work, which I think is most types of video file. You can put images and video in the same folder. But it may be convenient to separate them and have separate [[directory]] entries in the dataset config file.