INNER CODE UNIT · Python

GptDialogueDataset

Koziev/NLP_Datasets · Conversations/Data/run_chitchat_query.py:17

class GptDialogueDataset(torch.utils.data.Dataset):
    def __init__(self, tokenizer, file_path, block_size):
        self.examples = []
        self.transform = None
        with open(file_path, encoding="utf-8") as f:
            chunks = f.read().split('\n\n')  # пустая строка (или две) - разделитель диалогов
            for chunk in chunks:
                if len(chunk) > 0:
                    tokenized_chunk = tokenizer.convert_tokens_to_ids(tokenizer.tokenize(chunk.strip()))
                    l = len(tokenized_chunk)
                    if l < block_size:
                        self.examples.append(tokenizer.build_inputs_with_special_tokens(tokenized_chunk))
                    else:
                        while tokenized_chunk:
                            self.examples.append(tokenizer.build_inputs_with_special_tokens(tokenized_chunk[:block_size]))
                            tokenized_chunk = tokenized_chunk[block_size:]

    def __len__(self):

View source record →