OOM issue during training

[my3636]

Hi, thanks for releasing this great project!

I get an out-of-memory (OOM) when running run_scripts/padt_ovd_3b_sft.sh with:

• 2 × RTX 4090D (24GB)
• per_device_train_batch_size = 1

Could you please clarify:

Is the LLM (Qwen2.5-VL-3B-Instruct) fine-tuned (fully/partially) or frozen during training?

What hyperparameters should be adjusted to make this runnable, or what is the minimum GPU memory / GPU requirement?

Thanks in advance!