|
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 = get_tests_dir("fixtures/test_sentencepiece.model") |
|
spm_model_file_with_bytefallback = 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), |
|
) |
|
|
