| # test tokenizer encode & decode consistency | |
| from transformers import AutoTokenizer | |
| tokenizer = AutoTokenizer.from_pretrained('/apdcephfs/share_1502809/shaneshu/tokenizer_exp/other_tokenizer_vocab/hy', local_files_only=True, trust_remote_code=True) | |
| test_data = [line.strip() for line in open('/apdcephfs/share_1502809/shaneshu/tokenizer_exp/data/test.txt', 'r').readlines()] | |
| num_origi_len = 0 | |
| num_token_len = 0 | |
| for d in test_data: | |
| a = tokenizer.encode(d) | |
| num_origi_len += len(d) | |
| num_token_len += len(a) | |
| b = tokenizer.decode(a) | |
| assert b == d, f"encode & decode not consistent: {d} vs {b}" | |
| print(f" original length: {num_origi_len}") | |
| print(f" token length: {num_token_len}") | |