app.py

import torch
import gradio as gr
from utils import create_vocab, setup_seed
from dataset_mlm import  get_paded_token_idx_gen, add_tokens_to_vocab
setup_seed(4)
device = torch.device("cpu")
vocab_mlm = create_vocab()
vocab_mlm = add_tokens_to_vocab(vocab_mlm)
save_path = 'mlm-model-27.pt'
model = torch.load(save_path)
model = model.to(device)

def CTXGen(X1, X2, X3, top_k):
    predicted_token_probability_all = []
    model.eval()
    topk = []
    with torch.no_grad():
        new_seq = None
        seq = [f"{X1}|{X2}|{X3}|||"]
        vocab_mlm.token_to_idx["X"] = 4
        padded_seq, _, idx_msa, _ = get_paded_token_idx_gen(vocab_mlm, seq, new_seq)
        idx_msa = torch.tensor(idx_msa).unsqueeze(0).to(device)
        mask_positions = [i for i, token in enumerate(padded_seq) if token == "X"]
        if not mask_positions:
            raise ValueError("Nothing found in the sequence to predict.")

        for mask_position in mask_positions:
            padded_seq[mask_position] = "[MASK]"
            input_ids = vocab_mlm.__getitem__(padded_seq)
            input_ids = torch.tensor([input_ids]).to(device)
            logits = model(input_ids, idx_msa)
            mask_logits = logits[0, mask_position, :]
            predicted_token_probability, predicted_token_id = torch.topk((torch.softmax(mask_logits, dim=-1)), k=top_k)
            topk.append(predicted_token_id)
            predicted_token = vocab_mlm.idx_to_token[predicted_token_id[0].item()]
            predicted_token_probability_all.append(predicted_token_probability[0].item())
            padded_seq[mask_position] = predicted_token

        cls_pos = vocab_mlm.to_tokens(list(topk[0]))
        Topk = cls_pos
        if X1 != "X":
            Subtype = X1
            Potency = padded_seq[2],predicted_token_probability_all[0]
        elif X2 != "X":
            Subtype = padded_seq[1],predicted_token_probability_all[0]
            Potency = X2
        else:
            Subtype = padded_seq[1],predicted_token_probability_all[0]
            Potency = padded_seq[2],predicted_token_probability_all[1]
    return Subtype, Potency, Topk

iface = gr.Interface(fn=CTXGen,
                     inputs=["text", "text", "text", "text"],
                    outputs= ["text", "text", "text"])
iface.launch()