Skip to content

[from_pretrained] Allow tokenizer_type ≠ model_type #6995

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Sep 9, 2020
Merged

[from_pretrained] Allow tokenizer_type ≠ model_type #6995

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
1 change: 1 addition & 0 deletions src/transformers/configuration_utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -190,6 +190,7 @@ def __init__(self, **kwargs):
self.num_labels = kwargs.pop("num_labels", 2)

# Tokenizer arguments TODO: eventually tokenizer and models should share the same config
self.tokenizer_class = kwargs.pop("tokenizer_class", None)
self.prefix = kwargs.pop("prefix", None)
self.bos_token_id = kwargs.pop("bos_token_id", None)
self.pad_token_id = kwargs.pop("pad_token_id", None)
Expand Down
1 change: 1 addition & 0 deletions src/transformers/testing_utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -14,6 +14,7 @@

SMALL_MODEL_IDENTIFIER = "julien-c/bert-xsmall-dummy"
DUMMY_UNKWOWN_IDENTIFIER = "julien-c/dummy-unknown"
DUMMY_DIFF_TOKENIZER_IDENTIFIER = "julien-c/dummy-diff-tokenizer"
# Used to test Auto{Config, Model, Tokenizer} model_type detection.


Expand Down
11 changes: 11 additions & 0 deletions src/transformers/tokenization_auto.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -217,6 +217,17 @@ def from_pretrained(cls, pretrained_model_name_or_path, *inputs, **kwargs):
return BertJapaneseTokenizer.from_pretrained(pretrained_model_name_or_path, *inputs, **kwargs)

use_fast = kwargs.pop("use_fast", False)

if config.tokenizer_class is not None:
if use_fast and not config.tokenizer_class.endswith("Fast"):
tokenizer_class_candidate = f"{config.tokenizer_class}Fast"
else:
tokenizer_class_candidate = config.tokenizer_class
tokenizer_class = globals().get(tokenizer_class_candidate)
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Might be cleaner to use some of our internal list/dict containing all tokenizers, just in case there are weird things in the namespace of some users.

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yes I was wondering about that. I was wondering if by using globals() someone could even use a tokenizer that's not in the library, but I don't think so, as globals is actually locals in this scope/file if I understand correctly.

if tokenizer_class is None:
raise ValueError("Tokenizer class {} does not exist or is not currently imported.")
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The content of the {} is missing? Or there is a magic somewhere in ValueError which fills this?

Copy link
Member Author

@julien-c julien-c Sep 8, 2020 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

oops no it's missing

return tokenizer_class.from_pretrained(pretrained_model_name_or_path, *inputs, **kwargs)

for config_class, (tokenizer_class_py, tokenizer_class_fast) in TOKENIZER_MAPPING.items():
if isinstance(config, config_class):
if tokenizer_class_fast and use_fast:
Expand Down
16 changes: 15 additions & 1 deletion tests/test_tokenization_auto.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -27,7 +27,13 @@
RobertaTokenizer,
RobertaTokenizerFast,
)
from transformers.testing_utils import DUMMY_UNKWOWN_IDENTIFIER, SMALL_MODEL_IDENTIFIER # noqa: F401
from transformers.configuration_auto import AutoConfig
from transformers.configuration_roberta import RobertaConfig
from transformers.testing_utils import (
DUMMY_DIFF_TOKENIZER_IDENTIFIER,
DUMMY_UNKWOWN_IDENTIFIER,
SMALL_MODEL_IDENTIFIER,
)
from transformers.tokenization_auto import TOKENIZER_MAPPING


Expand Down Expand Up @@ -56,6 +62,14 @@ def test_tokenizer_from_model_type(self):
self.assertIsInstance(tokenizer, (RobertaTokenizer, RobertaTokenizerFast))
self.assertEqual(tokenizer.vocab_size, 20)

def test_tokenizer_from_tokenizer_class(self):
config = AutoConfig.from_pretrained(DUMMY_DIFF_TOKENIZER_IDENTIFIER)
self.assertIsInstance(config, RobertaConfig)
# Check that tokenizer_type ≠ model_type
tokenizer = AutoTokenizer.from_pretrained(DUMMY_DIFF_TOKENIZER_IDENTIFIER, config=config)
self.assertIsInstance(tokenizer, (BertTokenizer, BertTokenizerFast))
self.assertEqual(tokenizer.vocab_size, 12)

def test_tokenizer_identifier_with_correct_config(self):
for tokenizer_class in [BertTokenizer, BertTokenizerFast, AutoTokenizer]:
tokenizer = tokenizer_class.from_pretrained("wietsedv/bert-base-dutch-cased")
Expand Down