How to do fine tuning with distilled data

[ryosuzaki]

I would like to do fine tuning with distilled data.
The finetune_demo.sh does not seem to show how to do this.
I already have .cif files of distillation data.
I'm a newbie, so sorry if I'm looking in the wrong place.

[zhangyuxuann]

Hi @ryosuzaki you can refer to #61 first.

[ryosuzaki]

Thank you @zhangyuxuann. I read it.
After running python3 scripts/prepare_training_data.py -i [input_path] -o [output_csv] -b [output_dir] -c [cluster_txt] -n [num_cpu] -d and creating a data set, How should I modify finetune_demo.sh to specify the dataset?

[zhangyuxuann]

@ryosuzaki you can register a dataset like

Protenix/configs/configs_data.py

Line 126 in 1de0b3b

"weightedPDB_before2109_wopb_nometalc_0925": {

and modify the finetune_demo.sh like #120 (comment)