librarian-bots/model-card-sentences

id stringlengths 4 117	sentence stringlengths 1 134k
bert-base-uncased	340M
bert-base-uncased	English
bert-base-uncased	bert-large-cased-whole-word-masking
bert-base-uncased	340
bert-base-uncased	M
bert-base-uncased	English
bert-base-uncased	Intended uses & limitations
bert-base-uncased	You can use the raw model for either masked language modeling or next sentence prediction, but it's mostly intended to be fine-tuned on a downstream task.
bert-base-uncased	See the model hub to look for fine-tuned versions of a task that interests you.
bert-base-uncased	Note that this model is primarily aimed at being fine-tuned on tasks that use the whole sentence (potentially masked) to make decisions, such as sequence classification, token classification or question answering.
bert-base-uncased	For tasks such as text generation you should look at model like GPT2.
bert-base-uncased	How to use
bert-base-uncased	You can use this model directly with a pipeline for masked language modeling:
bert-base-uncased	>>> from transformers import pipeline >>> unmasker = pipeline('fill-mask', model='bert-base-uncased') >>>
bert-base-uncased	unmasker("Hello I'm a [MASK] model.")
bert-base-uncased	[{'sequence': "[CLS] hello i'm a fashion model.
bert-base-uncased	[SEP]", 'score': 0.1073106899857521, 'token': 4827, 'token_str': 'fashion'}, {'sequence': "[CLS] hello i'm a role model.
bert-base-uncased	[SEP]", 'score': 0.08774490654468536, 'token': 2535, 'token_str': 'role'}, {'sequence': "[CLS] hello i'm a new model.
bert-base-uncased	[SEP]", 'score': 0.05338378623127937, 'token': 2047, 'token_str': 'new'}, {'sequence': "[CLS] hello i'm a super model.
bert-base-uncased	[SEP]", 'score': 0.04667217284440994, 'token': 3565, 'token_str': 'super'}, {'sequence': "[CLS] hello i'm a fine model.
bert-base-uncased	[SEP]", 'score': 0.027095865458250046, 'token': 2986, 'token_str': 'fine'}]
bert-base-uncased	Here is how to use this model to get the features of a given text in PyTorch:
bert-base-uncased	from transformers import BertTokenizer, BertModel tokenizer = BertTokenizer.from_pretrained('bert-base-uncased') model = BertModel.from_pretrained("bert-base-uncased") text = "Replace me by any text you'd like."
bert-base-uncased	encoded_input = tokenizer(text, return_tensors='pt') output = model(**encoded_input)
bert-base-uncased	and in TensorFlow:
bert-base-uncased	from transformers import BertTokenizer, TFBertModel tokenizer = BertTokenizer.from_pretrained('bert-base-uncased') model = TFBertModel.from_pretrained("bert-base-uncased") text = "Replace me by any text you'd like."
bert-base-uncased	encoded_input = tokenizer(text, return_tensors='tf') output = model(encoded_input)
bert-base-uncased	Limitations and bias
bert-base-uncased	Even if the training data used for this model could be characterized as fairly neutral, this model can have biased predictions:
bert-base-uncased	>>> from transformers import pipeline >>> unmasker = pipeline('fill-mask', model='bert-base-uncased') >>>
bert-base-uncased	unmasker("The man worked as a [MASK].")
bert-base-uncased	[{'sequence': '[CLS] the man worked as a carpenter.
bert-base-uncased	[SEP]', 'score': 0.09747550636529922, 'token': 10533, 'token_str': 'carpenter'}, {'sequence': '[CLS] the man worked as a waiter.
bert-base-uncased	[SEP]', 'score': 0.0523831807076931, 'token': 15610, 'token_str': 'waiter'}, {'sequence': '[CLS] the man worked as a barber.
bert-base-uncased	[SEP]', 'score': 0.04962705448269844, 'token': 13362, 'token_str': 'barber'}, {'sequence': '
bert-base-uncased	[CLS] the man worked as a mechanic.
bert-base-uncased	[SEP]', 'score': 0.03788609802722931, 'token': 15893, 'token_str': 'mechanic'}, {'sequence': '[CLS] the man worked as a salesman.
bert-base-uncased	[SEP]', 'score': 0.037680890411138535, 'token': 18968, 'token_str': 'salesman'}] >>>
bert-base-uncased	unmasker("The woman worked as a [MASK].")
bert-base-uncased	[{'sequence': '[CLS] the woman worked as a nurse.
bert-base-uncased	[SEP]', 'score': 0.21981462836265564, 'token': 6821, 'token_str': 'nurse'}, {'sequence': '[CLS] the woman worked as a waitress.
bert-base-uncased	[SEP]', 'score': 0.1597415804862976, 'token': 13877, 'token_str': 'waitress'}, {'sequence': '[CLS] the woman worked as a maid.
bert-base-uncased	[SEP]', 'score': 0.1154729500412941, 'token': 10850, 'token_str': 'maid'}, {'sequence': '[CLS] the woman worked as a prostitute.
bert-base-uncased	[SEP]', 'score': 0.037968918681144714, 'token': 19215, 'token_str': 'prostitute'}, {'sequence': '[CLS] the woman worked as a cook.
bert-base-uncased	[SEP]', 'score': 0.03042375110089779, 'token': 5660, 'token_str': 'cook'}]
bert-base-uncased	This bias will also affect all fine-tuned versions of this model.
bert-base-uncased	Training data
bert-base-uncased	The BERT model was pretrained on BookCorpus, a dataset consisting of 11,038 unpublished books and English Wikipedia (excluding lists, tables and headers).
bert-base-uncased	Training procedure
bert-base-uncased	Preprocessing
bert-base-uncased	The texts are lowercased and tokenized using WordPiece and a vocabulary size of 30,000.
bert-base-uncased	The inputs of the model are then of the form:
bert-base-uncased	[CLS] Sentence A
bert-base-uncased	[SEP] Sentence B
bert-base-uncased	[SEP]
bert-base-uncased	With probability 0.5, sentence A and sentence B correspond to two consecutive sentences in the original corpus, and in the other cases, it's another random sentence in the corpus.
bert-base-uncased	Note that what is considered a sentence here is a consecutive span of text usually longer than a single sentence.
bert-base-uncased	The only constrain is that the result with the two "sentences" has a combined length of less than 512 tokens.
bert-base-uncased	The details of the masking procedure for each sentence are the following:
bert-base-uncased	15% of the tokens are masked.
bert-base-uncased	In 80% of the cases, the masked tokens are replaced by [MASK].
bert-base-uncased	In 10% of the cases, the masked tokens are replaced by a random token (different) from the one they replace.
bert-base-uncased	In the 10% remaining cases, the masked tokens are left as is.
bert-base-uncased	Pretraining
bert-base-uncased	The model was trained on 4 cloud TPUs in Pod configuration (16 TPU chips total) for one million steps with a batch size of 256.
bert-base-uncased	The sequence length was limited to 128 tokens for 90% of the steps and 512 for the remaining 10%.
bert-base-uncased	The optimizer used is Adam with a learning rate of 1e-4, \(\beta_{1} = 0.9\) and \(\beta_{2} = 0.999\), a weight decay of 0.01, learning rate warmup for 10,000 steps and linear decay of the learning rate after.
bert-base-uncased	Evaluation results
bert-base-uncased	When fine-tuned on downstream tasks, this model achieves the following results:
bert-base-uncased	Glue test results:
bert-base-uncased	Task
bert-base-uncased	MNLI-(m
bert-base-uncased	/mm)
bert-base-uncased	QQP
bert-base-uncased	QNLI
bert-base-uncased	SST-2
bert-base-uncased	CoLA
bert-base-uncased	STS-B
bert-base-uncased	MRPC
bert-base-uncased	RTE
bert-base-uncased	Average
bert-base-uncased	84.6/83.4
bert-base-uncased	71.2
bert-base-uncased	90.5
bert-base-uncased	93.5
bert-base-uncased	52.1
bert-base-uncased	85.8
bert-base-uncased	88.9
bert-base-uncased	66.4
bert-base-uncased	79.6
bert-base-uncased	BibTeX entry and citation info
bert-base-uncased	@article{DBLP:journals/corr/abs-1810-04805, author = {Jacob Devlin and Ming{-}Wei Chang and Kenton Lee and Kristina Toutanova}, title = {{BERT:}
bert-base-uncased	Pre-training of Deep Bidirectional Transformers for Language Understanding}, journal = {CoRR}, volume = {abs/1810.04805}, year = {2018}, url = {http://arxiv.org/abs/1810.04805}, archivePrefix = {arXiv}, eprint = {1810.04805}, timestamp = {Tue, 30 Oct 2018 20:39:56 +0100}, biburl = {https://dblp.org/rec/journals/corr/abs-1810-04805.bib}, bibsource = {dblp computer science bibliography, https://dblp.org} }
timm/mobilenetv3_large_100.ra_in1k	Model card for mobilenetv3_large_100.ra_in1k
timm/mobilenetv3_large_100.ra_in1k	A MobileNet-v3 image classification model.
timm/mobilenetv3_large_100.ra_in1k	Trained on ImageNet-1k in timm using recipe template described below.
timm/mobilenetv3_large_100.ra_in1k	Recipe details:
timm/mobilenetv3_large_100.ra_in1k	RandAugment RA recipe.
timm/mobilenetv3_large_100.ra_in1k	Inspired by and evolved from EfficientNet RandAugment recipes.
timm/mobilenetv3_large_100.ra_in1k	Published as B recipe in ResNet Strikes Back.

bert-base-uncased

340M

bert-base-uncased

English

bert-base-uncased

bert-large-cased-whole-word-masking

bert-base-uncased

340

bert-base-uncased

M

bert-base-uncased

English

bert-base-uncased

Intended uses & limitations

bert-base-uncased

You can use the raw model for either masked language modeling or next sentence prediction, but it's mostly intended to be fine-tuned on a downstream task.

bert-base-uncased

See the model hub to look for fine-tuned versions of a task that interests you.

bert-base-uncased

Note that this model is primarily aimed at being fine-tuned on tasks that use the whole sentence (potentially masked) to make decisions, such as sequence classification, token classification or question answering.

bert-base-uncased

For tasks such as text generation you should look at model like GPT2.

bert-base-uncased

How to use

bert-base-uncased

You can use this model directly with a pipeline for masked language modeling:

bert-base-uncased

>>> from transformers import pipeline >>> unmasker = pipeline('fill-mask', model='bert-base-uncased') >>>

bert-base-uncased

unmasker("Hello I'm a [MASK] model.")

bert-base-uncased

[{'sequence': "[CLS] hello i'm a fashion model.

bert-base-uncased

[SEP]", 'score': 0.1073106899857521, 'token': 4827, 'token_str': 'fashion'}, {'sequence': "[CLS] hello i'm a role model.

bert-base-uncased

[SEP]", 'score': 0.08774490654468536, 'token': 2535, 'token_str': 'role'}, {'sequence': "[CLS] hello i'm a new model.

bert-base-uncased

[SEP]", 'score': 0.05338378623127937, 'token': 2047, 'token_str': 'new'}, {'sequence': "[CLS] hello i'm a super model.

bert-base-uncased

[SEP]", 'score': 0.04667217284440994, 'token': 3565, 'token_str': 'super'}, {'sequence': "[CLS] hello i'm a fine model.

bert-base-uncased

[SEP]", 'score': 0.027095865458250046, 'token': 2986, 'token_str': 'fine'}]

bert-base-uncased

Here is how to use this model to get the features of a given text in PyTorch:

bert-base-uncased

from transformers import BertTokenizer, BertModel tokenizer = BertTokenizer.from_pretrained('bert-base-uncased') model = BertModel.from_pretrained("bert-base-uncased") text = "Replace me by any text you'd like."

bert-base-uncased

encoded_input = tokenizer(text, return_tensors='pt') output = model(**encoded_input)

bert-base-uncased

and in TensorFlow:

bert-base-uncased

from transformers import BertTokenizer, TFBertModel tokenizer = BertTokenizer.from_pretrained('bert-base-uncased') model = TFBertModel.from_pretrained("bert-base-uncased") text = "Replace me by any text you'd like."

bert-base-uncased

encoded_input = tokenizer(text, return_tensors='tf') output = model(encoded_input)

bert-base-uncased

Limitations and bias

bert-base-uncased

Even if the training data used for this model could be characterized as fairly neutral, this model can have biased predictions:

bert-base-uncased

>>> from transformers import pipeline >>> unmasker = pipeline('fill-mask', model='bert-base-uncased') >>>

bert-base-uncased

unmasker("The man worked as a [MASK].")

bert-base-uncased

[{'sequence': '[CLS] the man worked as a carpenter.

bert-base-uncased

[SEP]', 'score': 0.09747550636529922, 'token': 10533, 'token_str': 'carpenter'}, {'sequence': '[CLS] the man worked as a waiter.

bert-base-uncased

[SEP]', 'score': 0.0523831807076931, 'token': 15610, 'token_str': 'waiter'}, {'sequence': '[CLS] the man worked as a barber.

bert-base-uncased

[SEP]', 'score': 0.04962705448269844, 'token': 13362, 'token_str': 'barber'}, {'sequence': '

bert-base-uncased

[CLS] the man worked as a mechanic.

bert-base-uncased

[SEP]', 'score': 0.03788609802722931, 'token': 15893, 'token_str': 'mechanic'}, {'sequence': '[CLS] the man worked as a salesman.

bert-base-uncased

[SEP]', 'score': 0.037680890411138535, 'token': 18968, 'token_str': 'salesman'}] >>>

bert-base-uncased

unmasker("The woman worked as a [MASK].")

bert-base-uncased

[{'sequence': '[CLS] the woman worked as a nurse.

bert-base-uncased

[SEP]', 'score': 0.21981462836265564, 'token': 6821, 'token_str': 'nurse'}, {'sequence': '[CLS] the woman worked as a waitress.

bert-base-uncased

[SEP]', 'score': 0.1597415804862976, 'token': 13877, 'token_str': 'waitress'}, {'sequence': '[CLS] the woman worked as a maid.

bert-base-uncased

[SEP]', 'score': 0.1154729500412941, 'token': 10850, 'token_str': 'maid'}, {'sequence': '[CLS] the woman worked as a prostitute.

bert-base-uncased

[SEP]', 'score': 0.037968918681144714, 'token': 19215, 'token_str': 'prostitute'}, {'sequence': '[CLS] the woman worked as a cook.

bert-base-uncased

[SEP]', 'score': 0.03042375110089779, 'token': 5660, 'token_str': 'cook'}]

bert-base-uncased

This bias will also affect all fine-tuned versions of this model.

bert-base-uncased

Training data

bert-base-uncased

The BERT model was pretrained on BookCorpus, a dataset consisting of 11,038 unpublished books and English Wikipedia (excluding lists, tables and headers).

bert-base-uncased

Training procedure

bert-base-uncased

Preprocessing

bert-base-uncased

The texts are lowercased and tokenized using WordPiece and a vocabulary size of 30,000.

bert-base-uncased

The inputs of the model are then of the form:

bert-base-uncased

[CLS] Sentence A

bert-base-uncased

[SEP] Sentence B

bert-base-uncased

[SEP]

bert-base-uncased

With probability 0.5, sentence A and sentence B correspond to two consecutive sentences in the original corpus, and in the other cases, it's another random sentence in the corpus.

bert-base-uncased

Note that what is considered a sentence here is a consecutive span of text usually longer than a single sentence.

bert-base-uncased

The only constrain is that the result with the two "sentences" has a combined length of less than 512 tokens.

bert-base-uncased

The details of the masking procedure for each sentence are the following:

bert-base-uncased

15% of the tokens are masked.

bert-base-uncased

In 80% of the cases, the masked tokens are replaced by [MASK].

bert-base-uncased

In 10% of the cases, the masked tokens are replaced by a random token (different) from the one they replace.

bert-base-uncased

In the 10% remaining cases, the masked tokens are left as is.

bert-base-uncased

Pretraining

bert-base-uncased

The model was trained on 4 cloud TPUs in Pod configuration (16 TPU chips total) for one million steps with a batch size of 256.

bert-base-uncased

The sequence length was limited to 128 tokens for 90% of the steps and 512 for the remaining 10%.

bert-base-uncased

The optimizer used is Adam with a learning rate of 1e-4, \(\beta_{1} = 0.9\) and \(\beta_{2} = 0.999\), a weight decay of 0.01, learning rate warmup for 10,000 steps and linear decay of the learning rate after.

bert-base-uncased

Evaluation results

bert-base-uncased

When fine-tuned on downstream tasks, this model achieves the following results:

bert-base-uncased

Glue test results:

bert-base-uncased

Task

bert-base-uncased

MNLI-(m

bert-base-uncased

/mm)

bert-base-uncased

QQP

bert-base-uncased

QNLI

bert-base-uncased

SST-2

bert-base-uncased

CoLA

bert-base-uncased

STS-B

bert-base-uncased

MRPC

bert-base-uncased

RTE

bert-base-uncased

Average

bert-base-uncased

84.6/83.4

bert-base-uncased

71.2

bert-base-uncased

90.5

bert-base-uncased

93.5

bert-base-uncased

52.1

bert-base-uncased

85.8

bert-base-uncased

88.9

bert-base-uncased

66.4

bert-base-uncased

79.6

bert-base-uncased

BibTeX entry and citation info

bert-base-uncased

@article{DBLP:journals/corr/abs-1810-04805, author = {Jacob Devlin and Ming{-}Wei Chang and Kenton Lee and Kristina Toutanova}, title = {{BERT:}

bert-base-uncased

Pre-training of Deep Bidirectional Transformers for Language Understanding}, journal = {CoRR}, volume = {abs/1810.04805}, year = {2018}, url = {http://arxiv.org/abs/1810.04805}, archivePrefix = {arXiv}, eprint = {1810.04805}, timestamp = {Tue, 30 Oct 2018 20:39:56 +0100}, biburl = {https://dblp.org/rec/journals/corr/abs-1810-04805.bib}, bibsource = {dblp computer science bibliography, https://dblp.org} }