Failed to reproduce the results of predicting 6600AA long protein

Hi,

Thanks for sharing the good project. I have some issues related to FoldAcc on UniFold. I tried reproducing the results of predicting 6600 AA long protein which is mentioned in [this link](https://mp.weixin.qq.com/s/oSZ2hmWifiN5fIWRST2Dtw).

However, it failed with OOM error no matter what parameters I have set (e.g., **model_name=multimer_ft/multimer_af2, model_params=multimer.unifold.pt/AF2 original params, use one card A100 80G GPU/DAP=8, enable_low_mem=True, dtype=torch.bfloat16**). The details are attached as below.

- OOM error

- parameter settings

The input protein sequence is **F2ULY7**. Would you please provide any detailed guidance or running demo on how to reproduce the 6600AA long protein?

Start with the linked 6600 AA FoldAcc report and the UniFold prediction run for sequence F2ULY7, using the model, memory, device, DAP, and dtype settings listed in the issue. Compare the resulting OOM log with the reported run and determine whether the reproduction guidance or a runnable demo can be provided; done means documenting verified settings or explaining the limitation.

Written by the indexing model from the issue text.