Multilingual BERT

The Multilingual BERT block implements the BERT—​Bidirectional Encoder Representations from Transformers—​network in its base size, as published in BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding.

BERT pushes the state of the art in Natural Language Processing by combining two powerful technologies:

  • It is based on a deep Transformer encoder network, a type of network that can process long texts efficiently by using self-attention.

  • It is bidirectional, meaning that it uses the whole text passage to understand the meaning of each word.

Please note that datasets, machine-learning models, weights, topologies, research papers and other content, including open source software, (collectively referred to as “Content”) provided and/or suggested by Peltarion for use in the Platform and otherwise, may be subject to separate third party terms of use or license terms. You are solely responsible for complying with the applicable terms. Peltarion makes no representations or warranties about Content. You expressly relieve us from any and all liability, loss or risk arising (directly or indirectly) from Your use of any third party content.

What’s more, the original authors have released pre-trained weights, so that you can use it with minimal training work.

Why use a multilingual model?

A multilingual model allows you to deploy a single model able to work with any of the 100 known languages.

More than a simple convenience, multilingual models often perform better than monolingual models.
One reason is that the training data available is generally more limited in any single language. In addition, many languages share common patterns that the model can pick up more easily when it is trained with a variety of languages.


The input of the Multilingual BERT encoder must come from a BERT Tokenizer block.

The tokenizer must use Multilingual cased as Vocabulary, so that the tokenized numerical values are compatible with the Multilingual BERT encoder block.


The Multilingual BERT encoder returns the so-called CLS output. This output is a vector that can be passed to other blocks to perform regression or classification.

BERT Structure

The BERT Encoder block implements the base version of the BERT network. It is composed of 12 successive transformer layers, each having 12 attention heads.
The total number of parameters is 110 million.

BERT Encoder
Figure 1. Structure of BERT

Every token in the input of the block is first embedded into a learned 768-long embedding vector.

Each embedding vector is then transformed progressively every time it traverses one of the BERT Encoder layers:

  • Through linear projections, every embedding vector creates a triplet of 64-long vectors, called the key, query, and value vectors

  • The key, query, and value vectors from all the embeddings pass through a self-attention head, which outputs one 64-long vector for each input triplet.
    Every output vector from the self-attention head is a function of the whole input sequence, which is what makes BERT context-aware.

  • A single embedding vector uses different linear projections to create 12 unique triplets of key, query, and value vectors, which all go through their own self-attention head.
    This allows each self-attention head to focus on different aspects of how the tokens interact with each other.

  • The output from all the self-attention heads are first concatenated together, then they go through another linear projection and a feed-forward layer, which helps to utilize deep non-linearity. Residual connections from previous states are also used to increase robustness.

The result is a sequence of transformed embedding vectors, which are sent through the same layer structure 11 more times.

After the 12th encoding layer, the embedding vectors have been transformed to contain more accurate information about each token. This block returns only the first one (corresponding to the [CLS] token), which is often sufficient for classification tasks.

BERT Tokenizer

The BERT tokenizer converts plain text into a sequence of numerical values, which AI models love to handle.*

How does text tokenization work?

  1. The tokenizer splits the input text into small pieces, called tokens.
    There can be more tokens than words if parts of a word (like prefixes and suffixes) are more common than the word itself.

  2. The Sequence length is enforced by truncating or padding the sequence of tokens.

    Token padding
  3. Special tokens required by the Multilingual BERT encoder and English BERT encoder blocks are added, and every token is then replaced with an integer value.

    Token ID
  4. The sequence of integers is ready to be processed by one of the language processing blocks.

The Tokenizer block uses WordPiece under the hood.
This means that it can process input text features written in over 100 languages, and be directly connected to a Multilingial BERT Encoder or English BERT Encoder block for advanced Natural Language Processing.


Sequence length: The total number of tokens kept in the sequence. It’s necessary to fix the sequence length, since models require fixed size inputs.
Minimum: 3
Maximum: 512

If the text input is longer than the Sequence length, the end of the text will be ignored. If the text input is smaller, the sequence will be padded with PAD tokens.

Choose a length that matches your typical text size to utilize all the data while avoiding unnecessary calculations on the padding tokens.

Token paddinger
Figure 2. A Sequence length of 5 causes long sentences to be truncated, and short sentences to be padded.

Trainable: whether we want the training algorithm to change the value of the weights during training. In some cases, one will want to keep parts of the network static, e.g., when using the encoder part of an autoencoder as preprocessing for another model.

Was this page helpful?
Yes No