ProBEiT PyTorch model weights and datasets

ProBEiT is a multimodal protein representation model that jointly encodes amino-acid sequences and tokenized structural information. It uses two ESM-2 streams followed by a BEiT3-style multiway encoder and task-specific heads for protein property prediction and functional annotation. Training loads a pretrained ProBEiT state dict from `--model_path` (default `model/model_parameters/model.pth`) and copies over every parameter whose name and shape match the current task model. This checkpoint is required: the path is loaded with `torch.load`, so a missing file will raise an error. Due to GitHub file size limitations, the full datasets are hosted on Zenodo.