SUMIN/HuggingFace_transformer
1
0
Fork 0
Code Issues Pull Requests Actions 7 Packages Projects Releases Wiki Activity

[Core Tokenization] Support a fix for spm fast models (#26678)

Browse Source 
* fix

* last attempt

* current work

* fix forward compatibility

* save all special tokens

* current state

* revert additional changes

* updates

* remove tokenizer.model

* add a test and the fix

* nit

* revert one more break

* fix typefield issue

* quality

* more tests

* fix fields for FC

* more nits?

* new additional changes

* how

* some updates

* the fix

* where do we stand

* nits

* nits

* revert unrelated changes

* nits nits nits

* styling

* don't break llama just yet

* revert llama changes

* safe arg check

* fixup

* Add a test for T5

* Necessary changes

* Tests passing, added tokens need to not be normalized. If the added tokens are normalized, it will the stripping which seems to be unwanted for a normal functioning

* Add even more tests, when normalization is set to True (which does not work 😓 )

* Add even more tests, when normalization is set to True (which does not work 😓 )

* Update to main

* nits

* fmt

* more and more test

* comments

* revert change as tests are failing

* make the test more readble

* nits

* refactor the test

* nit

* updates

* simplify

* style

* style

* style convert slow

* Update src/transformers/convert_slow_tokenizer.py
This commit is contained in:
Arthur
2024-01-18 12:31:54 +01:00
committed by GitHub
parent a1668cc72e
commit 8189977885
2 changed files with 47 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

35
tests/models/t5/test_tokenization_t5.py
View File

@@ -424,6 +424,41 @@ class T5TokenizationTest(TokenizerTesterMixin, unittest.TestCase):
self.assertEqual(tokens, [])
self.assertEqual(tokens, tokenizer.sp_model.encode("▁", out_type=str))
def test_fast_slow_edge_cases(self):
# We are testing spaces before and spaces after special tokens + space transformations
slow_tokenizer = T5Tokenizer.from_pretrained("t5-base", legacy=False)
fast_tokenizer = T5TokenizerFast.from_pretrained("t5-base", legacy=False, from_slow=True)
slow_tokenizer.add_tokens(AddedToken("<new_token_test_>", rstrip=False, lstrip=False, normalized=False))
fast_tokenizer.add_tokens(AddedToken("<new_token_test_>", rstrip=False, lstrip=False, normalized=False))
edge_case = "Hey!<new_token_test_>. How</s>Hey <new_token_test_>!"
EXPECTED_SLOW = ["▁Hey", "!", "<new_token_test_>", ".", "▁How", "</s>", "He", "y", "<new_token_test_>", "!"] # fmt: skip
with self.subTest(f"slow {edge_case} normalized = False"):
self.assertEqual(slow_tokenizer.tokenize(edge_case), EXPECTED_SLOW)
with self.subTest(f"Fast {edge_case} normalized = False"):
self.assertEqual(fast_tokenizer.tokenize(edge_case), EXPECTED_SLOW)
hard_case = "Hey! <new_token_test_>. How</s> Hey <new_token_test_> ! . "
EXPECTED_SLOW = ["▁Hey", "!", "<new_token_test_>", ".", "▁How", "</s>", "▁Hey", "<new_token_test_>", "▁", "!", "▁", "."] # fmt: skip
with self.subTest(f"slow {edge_case} normalized = False"):
self.assertEqual(slow_tokenizer.tokenize(hard_case), EXPECTED_SLOW)
with self.subTest(f"fast {edge_case} normalized = False"):
self.assertEqual(fast_tokenizer.tokenize(hard_case), EXPECTED_SLOW)
fast_tokenizer = T5TokenizerFast.from_pretrained("t5-base", legacy=False, from_slow=True)
fast_tokenizer.add_tokens(AddedToken("<new_token_test_>", rstrip=False, lstrip=False, normalized=True))
# `normalized=True` is the default normalization scheme when adding a token. Normalize -> don't strip the space.
# the issue now is that our slow tokenizer should NOT strip the space if we want to simulate sentencepiece token addition.
EXPECTED_FAST = ["▁Hey", "!", "<new_token_test_>", ".", "▁How", "</s>", "He", "y", "▁", "<new_token_test_>", "!"] # fmt: skip
with self.subTest(f"fast {edge_case} normalized = True"):
self.assertEqual(fast_tokenizer.tokenize(edge_case), EXPECTED_FAST)
EXPECTED_FAST = ['▁Hey', '!', '▁', '<new_token_test_>', '.', '▁How', '</s>', '▁Hey','▁', '<new_token_test_>', '▁', '!', '▁', '.'] # fmt: skip
with self.subTest(f"fast {edge_case} normalized = False"):
self.assertEqual(fast_tokenizer.tokenize(hard_case), EXPECTED_FAST)
@require_sentencepiece
@require_tokenizers