westlake-repl/SaProt_35M_AF2_seqOnly

This model is provided to compare with official ESM-2 35M model. It only receives residue sequence but shares the same vocabulary with normal SaProt, which means all structure(3Di) tokens are marked as # during training.

Huggingface model

The following code shows how to load the model.

from transformers import EsmTokenizer, EsmForMaskedLM

model_path = "/your/path/to/SaProt_35M_AF2_seqOnly"
tokenizer = EsmTokenizer.from_pretrained(model_path)
model = EsmForMaskedLM.from_pretrained(model_path)

#################### Example ####################
device = "cuda"
model.to(device)

seq = "M#E#V#Q#L#V#Q#Y#K#"
tokens = tokenizer.tokenize(seq)
print(tokens)

inputs = tokenizer(seq, return_tensors="pt")
inputs = {k: v.to(device) for k, v in inputs.items()}

outputs = model(**inputs)
print(outputs.logits.shape)

"""
['M#', 'E#', 'V#', 'Q#', 'L#', 'V#', 'Q#', 'Y#', 'K#']
torch.Size([1, 11, 446])
"""