| import sys |
| import time |
| import inspect |
|
|
| from transformers import AutoTokenizer |
| from typing import Any |
| import numpy as np |
| from tqdm import tqdm |
|
|
| import json |
| import argparse |
| import os |
|
|
| def parse_args(): |
| parser = argparse.ArgumentParser(description="Finetune a transformers model on a causal language modeling task") |
| parser.add_argument( |
| "--source_file", |
| type=str, |
| ) |
| parser.add_argument( |
| "--max_length", |
| type=int, |
| default=512, |
| ) |
| parser.add_argument( |
| "--chunk_size", |
| type=int, |
| default=1024, |
| ) |
| parser.add_argument( |
| "--tokenizer_path", |
| type=str, |
| ) |
| args = parser.parse_args() |
| return args |
|
|
|
|
| def get_tokenizer(tokenizer_path): |
| tokenizer = tokenizer = AutoTokenizer.from_pretrained( |
| tokenizer_path, use_fast=not False, trust_remote_code=False |
| ) |
| |
| |
| return tokenizer |
|
|
|
|
| def convert_data_to_id(tokenizer: AutoTokenizer, data: Any): |
| input_ids = tokenizer.encode(data) |
| ids = input_ids |
| ids = np.array(ids, dtype=np.int32) |
| return ids |
|
|
| args = parse_args() |
|
|
| tokenizer = get_tokenizer(args.tokenizer_path) |
| infile = open(args.source_file, 'r', encoding='utf-8') |
| file_name, _ = os.path.splitext(os.path.basename(args.source_file)) |
|
|
| print("source file - ", args.source_file) |
| print('############ Start data reading ###########') |
|
|
| idx = 0 |
| max_length = args.max_length |
| chunk_size = args.chunk_size |
|
|
| token_ids = np.array([], dtype=np.int32) |
|
|
| with open(file_name+'_streaming_'+str(max_length)+'.jsonl', 'w') as f: |
| for line in infile: |
| idx += 1 |
| if idx % 10000 == 0: |
| print('Cur idx - ', idx) |
| try: |
| line = json.loads(line) |
| cur_texts = [] |
| if 'text' in line: |
| temp = line['text'] + "\n" |
| elif 'raw_content_lines' in line: |
| temp = "\n".join(line['raw_content_lines']) + "\n" |
| else: |
| print("error") |
| exit() |
| try: |
| token_id = convert_data_to_id(tokenizer, temp) |
| token_ids = np.concatenate((token_ids, token_id), dtype=np.int32) |
| except UnicodeDecodeError: |
| print('Error line - encoding: ', idx) |
|
|
| if len(token_ids) > max_length*chunk_size: |
| while len(token_ids) > max_length: |
| try: |
| temp_text = tokenizer.decode(token_ids[: max_length]) |
| temp_dic = {'text': temp_text} |
| f.write(json.dumps(temp_dic) + "\n") |
| token_ids = token_ids[max_length:] |
| except UnicodeDecodeError: |
| print('Error line - decoding: ', idx) |
| token_ids = token_ids[max_length:] |
| |
| except: |
| print("error source file - ", args.source_file) |
| print('Error line: ', idx) |
| continue |
|
|
| infile.close() |
