Spaces:
Running
Running
| from modeling import MT5ForConditionalGeneration | |
| from transformers import AutoTokenizer | |
| import os | |
| class ChemicalConverter: | |
| def __init__(self, mode: str): | |
| self.mode = mode | |
| model_directory = os.path.abspath("models") | |
| model_path = os.path.join(model_directory, mode) | |
| if mode == "SMILES2IUPAC": | |
| model_path = "knowledgator/SMILES2IUPAC-canonical-base" | |
| else: | |
| model_path = "knowledgator/IUPAC2SMILES-canonical-small" | |
| self.model = MT5ForConditionalGeneration.from_pretrained(model_path) | |
| self.smiles_tokenizer = AutoTokenizer.from_pretrained("knowledgator/SMILES-FAST-TOKENIZER") | |
| self.iupac_tokenizer = AutoTokenizer.from_pretrained("knowledgator/IUPAC-FAST-TOKENIZER") | |
| self.smiles_max_len = 128 | |
| self.iupac_max_len = 156 | |
| def convert(self, input): | |
| if self.mode == "SMILES2IUPAC": | |
| tokenizer = self.smiles_tokenizer | |
| reverse_tokenizer = self.iupac_tokenizer | |
| max_length = self.smiles_max_len | |
| else: | |
| tokenizer = self.iupac_tokenizer | |
| reverse_tokenizer = self.smiles_tokenizer | |
| max_length = self.iupac_max_len | |
| encoding = tokenizer(input, | |
| return_tensors='pt', | |
| padding="max_length", | |
| truncation=True, | |
| max_length=max_length) | |
| # Move the input tensor to GPU | |
| encoding = {key: value.to(self.model.device) for key, value in encoding.items()} | |
| # Generate names | |
| output = self.model.generate(input_ids=encoding['input_ids'], | |
| attention_mask=encoding['attention_mask'], | |
| max_new_tokens=156, | |
| num_beams=1, | |
| num_return_sequences=1) | |
| # Decode names | |
| output = [reverse_tokenizer.decode(ids, skip_special_tokens=True) for ids in output] | |
| return output[0] |