Out-of-memory for default config.

[chehx]

Many thanks for open-sourcing this great project.

Currently, I meet the out-of-memory error when training.

I use the default training config in stage3.py and I have 2 A100 80G.

However, it raises the error, but in report 1.1, it says the default config is for 80G memory usage.

Currently, when I use 480p with 48 frames, it takes around 73GB.

[JamesTensor]

I read report 1.1 and it does not state that only 80G of memory is required for training. Where did you see that?

[chehx]

I read report 1.1 and it does not state that only 80G of memory is required for training. Where did you see that?

https://github.com/hpcaitech/Open-Sora/issues/344#issuecomment-2102359347

Honestly, I saw this response. Did I misunderstand something?

[github-actions[bot]]

This issue is stale because it has been open for 7 days with no activity.

[chehx]

I found the problem!

When I wanna use the pre-trained weight via huggingface, it will load the config file:

https://huggingface.co/hpcai-tech/OpenSora-STDiT-v2-stage3/blob/main/config.json

where, the

"enable_flash_attn": false, "enable_layernorm_kernel": false,

is forbidden!

[github-actions[bot]]

This issue is stale because it has been open for 7 days with no activity.

[JThh]

I am gonna close this issue since it appears to have been resolved by the question owner.