mirror of
https://github.com/huggingface/transformers.git
synced 2025-07-03 21:00:08 +06:00
1025a9b742
* update proto sentencepiece model * Revert "update proto sentencepiece model" This reverts commitb07f671747. * add check * add test * Revert "Revert "update proto sentencepiece model"" This reverts commit46108257b8. * test for log level * test for log level 2 * warning at the warning level * clean * format * add explanation in docstring
37 lines
1.4 KiB
Python
37 lines
1.4 KiB
Python
import unittest
|
|
import warnings
|
|
from dataclasses import dataclass
|
|
|
|
from transformers.convert_slow_tokenizer import SpmConverter
|
|
from transformers.testing_utils import get_tests_dir
|
|
|
|
|
|
@dataclass
|
|
class FakeOriginalTokenizer:
|
|
vocab_file: str
|
|
|
|
|
|
class ConvertSlowTokenizerTest(unittest.TestCase):
|
|
def test_spm_converter_bytefallback_warning(self):
|
|
spm_model_file_without_bytefallback = f"{get_tests_dir()}/fixtures/test_sentencepiece.model"
|
|
spm_model_file_with_bytefallback = f"{get_tests_dir()}/fixtures/test_sentencepiece_with_bytefallback.model"
|
|
|
|
original_tokenizer_without_bytefallback = FakeOriginalTokenizer(vocab_file=spm_model_file_without_bytefallback)
|
|
|
|
with warnings.catch_warnings(record=True) as w:
|
|
_ = SpmConverter(original_tokenizer_without_bytefallback)
|
|
self.assertEqual(len(w), 0)
|
|
|
|
original_tokenizer_with_bytefallback = FakeOriginalTokenizer(vocab_file=spm_model_file_with_bytefallback)
|
|
|
|
with warnings.catch_warnings(record=True) as w:
|
|
_ = SpmConverter(original_tokenizer_with_bytefallback)
|
|
self.assertEqual(len(w), 1)
|
|
self.assertIn(
|
|
(
|
|
"The sentencepiece tokenizer that you are converting to a fast tokenizer uses the byte fallback option"
|
|
" which is not implemented in the fast tokenizers."
|
|
),
|
|
str(w[0].message),
|
|
)
|