torch.OutOfMemoryError when finetune

[mikeyimer]

Follow the finetuning commands,I tried setting max_image_size to 512, but still oom on 40g ,how much vram is needed to finetune? Are there any parameters can reduce the vram? thank you

[staoxiao]

You can use lora fine-tuning instead of full fine-tuning: https://github.com/VectorSpaceLab/OmniGen/blob/main/docs/fine-tuning.md#lora-fine-tuning

[whl0070179]

How much vram is needed to full fine-tuning？