Pretrained model on English language using a masked language modeling (MLM) objective. It was introduced in
this paper and first released in
this repository. This model, as all ALBERT models, is uncased: it does not make a difference
between english and English.
Disclaimer: The team releasing ALBERT did not write a model card for this model so this model card has been written by
the Hugging Face team.
Model description
ALBERT is a transformers model pretrained on a large corpus of English data in a self-supervised fashion. This means it
was pretrained on the raw texts only, with no humans labelling them in any way (which is why it can use lots of
publicly available data) with an automatic process to generate inputs and labels from those texts. More precisely, it
was pretrained with two objectives:
Masked language modeling (MLM): taking a sentence, the model randomly masks 15% of the words in the input then run
the entire masked sentence through the model and has to predict the masked words. This is different from traditional
recurrent neural networks (RNNs) that usually see the words one after the other, or from autoregressive models like
GPT which internally mask the future tokens. It allows the model to learn a bidirectional representation of the
sentence.
Sentence Ordering Prediction (SOP): ALBERT uses a pretraining loss based on predicting the ordering of two consecutive segments of text.
This way, the model learns an inner representation of the English language that can then be used to extract features
useful for downstream tasks: if you have a dataset of labeled sentences for instance, you can train a standard
classifier using the features produced by the ALBERT model as inputs.
ALBERT is particular in that it shares its layers across its Transformer. Therefore, all layers have the same weights. Using repeating layers results in a small memory footprint, however, the computational cost remains similar to a BERT-like architecture with the same number of hidden layers as it has to iterate through the same number of (repeating) layers.
This is the second version of the base model. Version 2 is different from version 1 due to different dropout rates, additional training data, and longer training. It has better results in nearly all downstream tasks.
This model has the following configuration:
12 repeating layers
128 embedding dimension
768 hidden dimension
12 attention heads
11M parameters
Intended uses & limitations
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. See the model hub to look for
fine-tuned versions on a task that interests you.
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. For tasks such as text
generation you should look at model like GPT2.
How to use
You can use this model directly with a pipeline for masked language modeling:
In tf_transformers
python
1from tf_transformers.models import AlbertModel
2from transformers import AlbertTokenizer
34tokenizer = AlbertTokenizer.from_pretrained('albert-xlarge-v2')5model = AlbertModel.from_pretrained("albert-xlarge-v2")67text ="Replace me by any text you'd like."8inputs_tf ={}9inputs = tokenizer(text, return_tensors='tf')101112inputs_tf["input_ids"]= inputs["input_ids"]13inputs_tf["input_type_ids"]= inputs["token_type_ids"]14inputs_tf["input_mask"]= inputs["attention_mask"]15outputs_tf = model(inputs_tf)
Training data
The ALBERT model was pretrained on BookCorpus, a dataset consisting of 11,038
unpublished books and English Wikipedia (excluding lists, tables and
headers).
Training procedure
Preprocessing
The texts are lowercased and tokenized using SentencePiece and a vocabulary size of 30,000. The inputs of the model are
then of the form:
[CLS] Sentence A [SEP] Sentence B [SEP]
Training
The ALBERT procedure follows the BERT setup.
The details of the masking procedure for each sentence are the following:
15% of the tokens are masked.
In 80% of the cases, the masked tokens are replaced by [MASK].
In 10% of the cases, the masked tokens are replaced by a random token (different) from the one they replace.
In the 10% remaining cases, the masked tokens are left as is.
Evaluation results
When fine-tuned on downstream tasks, the ALBERT models achieve the following results:
Average
SQuAD1.1
SQuAD2.0
MNLI
SST-2
RACE
V2
ALBERT-base
82.3
90.2/83.2
82.1/79.3
84.6
92.9
66.8
ALBERT-large
85.7
91.8/85.2
84.9/81.8
86.5
94.9
75.2
ALBERT-xlarge
87.9
92.9/86.4
87.9/84.1
87.9
95.4
80.7
ALBERT-xxlarge
90.9
94.6/89.1
89.8/86.9
90.6
96.8
86.8
V1
ALBERT-base
80.1
89.3/82.3
80.0/77.1
81.6
90.3
64.0
ALBERT-large
82.4
90.6/83.9
82.3/79.4
83.5
91.7
68.5
ALBERT-xlarge
85.5
92.5/86.1
86.1/83.1
86.4
92.4
74.8
ALBERT-xxlarge
91.0
94.8/89.3
90.2/87.4
90.8
96.9
86.5
BibTeX entry and citation info
bibtex
1@article{DBLP:journals/corr/abs-1909-11942,
2 author = {Zhenzhong Lan and
3 Mingda Chen and
4 Sebastian Goodman and
5 Kevin Gimpel and
6 Piyush Sharma and
7 Radu Soricut},
8 title = {{ALBERT:} {A} Lite {BERT} for Self-supervised Learning of Language
9 Representations},
10 journal = {CoRR},
11 volume = {abs/1909.11942},
12 year = {2019},
13 url = {http://arxiv.org/abs/1909.11942},
14 archivePrefix = {arXiv},
15 eprint = {1909.11942},
16 timestamp = {Fri, 27 Sep 2019 13:04:21 +0200},
17 biburl = {https://dblp.org/rec/journals/corr/abs-1909-11942.bib},
18 bibsource = {dblp computer science bibliography, https://dblp.org}
19}
