This model repo includes the checkopints for [ProteinDT paper](https://arxiv.org/abs/2302.04611).

- The pretraining is using ProteinCLAP-EBM-NCE on SwissProtCLAP dataset.
- The downstream adopts the TAPE dataset.

Now we conver:
- `pytorch_model_ss3.bin` for task SS-Q3.
- `pytorch_model_ss8.bin` for task SS-Q8.