该代码库基于以下项目进行适配：https://huggingface.co/chenxran/bart-smiles/tree/main

from transformers import AutoTokenizer, AutoModel, SequenceFeatureExtractor
import torch
from transformers import AutoTokenizer, AutoModel

smiles = "CCC(=O)"

tokenizer = AutoTokenizer.from_pretrained("./BARTSmiles/", add_prefix_space=True)
inputs = tokenizer(smiles, return_tensors="pt", return_token_type_ids=False, add_special_tokens=True)

model = AutoModel.from_pretrained('./BARTSmiles')
model.eval()

# 使用pipeline作为高级辅助工具
from transformers import pipeline

extractor = pipeline("feature-extraction", model=model, tokenizer=tokenizer)
result = extractor(smiles, return_tensors=True, tokenize_kwargs={'return_token_type_ids':False})

引用文献

@article{chilingaryan2022bartsmiles, title={Bartsmiles: 用于分子表征的生成式掩码语言模型}, author={奇林加良, 加亚内 and 塔莫扬, 霍夫汉内斯 and 特沃相, 阿尼 and 巴巴扬, 内莉 and 洪德卡良, 卢西内 and 汉巴尔祖米扬, 卡伦 and 纳沃扬, 扎文 and 哈恰特良, 赫兰特 and 阿加贾尼扬, 阿尔缅}, journal={arXiv预印本 arXiv:2211.16349}, year={2022} }