RoBERTa French base model (Uncased)

Prerequisites

transformers==4.19.2

Model architecture

This model uses RoBERTa base setttings except vocabulary size.

Tokenizer

Using BPE tokenizer with vocabulary size 50,000.

Training Data

• wiki40b/fr (French Wikipedia)
• Subset of CC-100/fr : Monolingual Datasets from Web Crawl Data

Usage

from transformers import pipeline

unmasker = pipeline('fill-mask', model='ClassCat/roberta-base-french')
unmasker("Je vais à la <mask>.")