Encodage par paires d'octets
en construction
Définition
XXXXXXX
Voir aussi segment, traitement automatique de la langue naturelle et Vocabulary (NLP)
Français
XXXXXXX
Anglais
Byte Pair Encoding
BPE
Byte Pair Encoding is a simple form of data compression algorithms and is one of the most widely used subword-tokenization algorithms. It replaces the most frequent pair of bytes of data with a new byte that was not contained int the initial dataset. In Natural Language Processing, BPE is used to represent large vocabulary with a small set of subword units and most common words are represented in the vocabulary as a single token.
It is used in all of GPT versions, RoBERTa, XML, FlauBERT and more.
Source
Contributeurs: Arianne
