Skip to content
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

Jump to bottom

Add XLMR and RoBERTa transforms as factory functions #2102

Merged
merged 3 commits into from
Mar 9, 2023
Merged

Add XLMR and RoBERTa transforms as factory functions #2102

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
f4a7a87
Add XLMR and RoBERTa transforms as classes
joecummings Mar 8, 2023
498036c
Remove unused import
joecummings Mar 8, 2023
e35cb8e
Add roberta transform and xlmr transform as factory methods
joecummings Mar 9, 2023
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
72 changes: 29 additions & 43 deletions torchtext/models/roberta/bundler.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -160,16 +160,35 @@ def encoderConf(self) -> RobertaEncoderConf:
return self._encoder_conf


XLMR_BASE_ENCODER = RobertaBundle(
_path=urljoin(_TEXT_BUCKET, "xlmr.base.encoder.pt"),
_encoder_conf=RobertaEncoderConf(vocab_size=250002),
transform=lambda: T.Sequential(
def xlmr_transform(truncate_length: int) -> Module:
"""Standard transform for XLMR models."""
return T.Sequential(
T.SentencePieceTokenizer(urljoin(_TEXT_BUCKET, "xlmr.sentencepiece.bpe.model")),
T.VocabTransform(load_state_dict_from_url(urljoin(_TEXT_BUCKET, "xlmr.vocab.pt"))),
T.Truncate(254),
T.Truncate(truncate_length),
T.AddToken(token=0, begin=True),
T.AddToken(token=2, begin=False),
),
)


def roberta_transform(truncate_length: int) -> Module:
"""Standard transform for RoBERTa models."""
return T.Sequential(
T.GPT2BPETokenizer(
encoder_json_path=urljoin(_TEXT_BUCKET, "gpt2_bpe_encoder.json"),
vocab_bpe_path=urljoin(_TEXT_BUCKET, "gpt2_bpe_vocab.bpe"),
),
T.VocabTransform(load_state_dict_from_url(urljoin(_TEXT_BUCKET, "roberta.vocab.pt"))),
T.Truncate(truncate_length),
T.AddToken(token=0, begin=True),
T.AddToken(token=2, begin=False),
)


XLMR_BASE_ENCODER = RobertaBundle(
_path=urljoin(_TEXT_BUCKET, "xlmr.base.encoder.pt"),
_encoder_conf=RobertaEncoderConf(vocab_size=250002),
transform=lambda: xlmr_transform(254),
)

XLMR_BASE_ENCODER.__doc__ = """
Expand All @@ -193,13 +212,7 @@ def encoderConf(self) -> RobertaEncoderConf:
_encoder_conf=RobertaEncoderConf(
vocab_size=250002, embedding_dim=1024, ffn_dimension=4096, num_attention_heads=16, num_encoder_layers=24
),
transform=lambda: T.Sequential(
T.SentencePieceTokenizer(urljoin(_TEXT_BUCKET, "xlmr.sentencepiece.bpe.model")),
T.VocabTransform(load_state_dict_from_url(urljoin(_TEXT_BUCKET, "xlmr.vocab.pt"))),
T.Truncate(510),
T.AddToken(token=0, begin=True),
T.AddToken(token=2, begin=False),
),
transform=lambda: xlmr_transform(510),
)

XLMR_LARGE_ENCODER.__doc__ = """
Expand All @@ -221,16 +234,7 @@ def encoderConf(self) -> RobertaEncoderConf:
ROBERTA_BASE_ENCODER = RobertaBundle(
_path=urljoin(_TEXT_BUCKET, "roberta.base.encoder.pt"),
_encoder_conf=RobertaEncoderConf(vocab_size=50265),
transform=lambda: T.Sequential(
T.GPT2BPETokenizer(
encoder_json_path=urljoin(_TEXT_BUCKET, "gpt2_bpe_encoder.json"),
vocab_bpe_path=urljoin(_TEXT_BUCKET, "gpt2_bpe_vocab.bpe"),
),
T.VocabTransform(load_state_dict_from_url(urljoin(_TEXT_BUCKET, "roberta.vocab.pt"))),
T.Truncate(254),
T.AddToken(token=0, begin=True),
T.AddToken(token=2, begin=False),
),
transform=lambda: roberta_transform(254),
)

ROBERTA_BASE_ENCODER.__doc__ = """
Expand Down Expand Up @@ -263,16 +267,7 @@ def encoderConf(self) -> RobertaEncoderConf:
num_attention_heads=16,
num_encoder_layers=24,
),
transform=lambda: T.Sequential(
T.GPT2BPETokenizer(
encoder_json_path=urljoin(_TEXT_BUCKET, "gpt2_bpe_encoder.json"),
vocab_bpe_path=urljoin(_TEXT_BUCKET, "gpt2_bpe_vocab.bpe"),
),
T.VocabTransform(load_state_dict_from_url(urljoin(_TEXT_BUCKET, "roberta.vocab.pt"))),
T.Truncate(510),
T.AddToken(token=0, begin=True),
T.AddToken(token=2, begin=False),
),
transform=lambda: roberta_transform(510),
)

ROBERTA_LARGE_ENCODER.__doc__ = """
Expand Down Expand Up @@ -302,16 +297,7 @@ def encoderConf(self) -> RobertaEncoderConf:
num_encoder_layers=6,
padding_idx=1,
),
transform=lambda: T.Sequential(
T.GPT2BPETokenizer(
encoder_json_path=urljoin(_TEXT_BUCKET, "gpt2_bpe_encoder.json"),
vocab_bpe_path=urljoin(_TEXT_BUCKET, "gpt2_bpe_vocab.bpe"),
),
T.VocabTransform(load_state_dict_from_url(urljoin(_TEXT_BUCKET, "roberta.vocab.pt"))),
T.Truncate(510),
T.AddToken(token=0, begin=True),
T.AddToken(token=2, begin=False),
),
transform=lambda: roberta_transform(510),
)

ROBERTA_DISTILLED_ENCODER.__doc__ = """
Expand Down