GPT2 Catalan small model Version 2 (Uncased)

Prerequisites

transformers==4.19.2

Model architecture

This model uses GPT2 base model settings, but the size of embedding dimensions are half the size of them.

Tokenizer

Using BPE tokenizer with vocabulary size 50,000.

Training Data

• wiki40b/ca (Catalan Wikipedia)
• Subset of oscar
• Subset of CC-100/ca : Monolingual Datasets from Web Crawl Data

Usage

from transformers import pipeline

unmasker = pipeline('fill-mask', model='ClassCat/gpt2-small-catalan-v2')
unmasker("Ell està una mica")