| def SimpleTokenizer(): | |
| import os | |
| from tokenizers import Tokenizer | |
| CURRENT_DIR = os.path.dirname(os.path.abspath(__file__)) | |
| TOKENIZER_DIR = os.path.join(CURRENT_DIR, "20B_tokenizer.json") | |
| tokenizer = Tokenizer.from_file(TOKENIZER_DIR) | |
| tokenizer.vocab_size = tokenizer.get_vocab_size(with_added_tokens=True) | |
| # vocab_size = len(tokenizer.get_vocab()) | |
| # vocab_size = tokenizer.vocab_size | |
| from transformers import AutoTokenizer | |
| tokenizer = AutoTokenizer.from_pretrained("EleutherAI/gpt-neox-20b") | |
| # "tokenizer_type": "HFTokenizer", # https://github.com/EleutherAI/gpt-neox/blob/v2.0/configs/20B.yml#L107 | |
| # tokenizer.vocab_size = tokenizer.get_vocab_size(with_added_tokens=True) | |