French T5 Abstractive Text Summarization

Version 1.0 (I will keep improving the model's performances.)

Version 2.0 is here! (with improved performances of course)

I trained the model on 13x more data than v1.

ROUGE-1: 44.5252

ROUGE-2: 22.652

ROUGE-L: 29.8866

Model description

This model is a T5 Transformers model (JDBN/t5-base-fr-qg-fquad) that was fine-tuned in french for abstractive text summarization.

How to use

from transformers import T5Tokenizer, T5ForConditionalGeneration
tokenizer = T5Tokenizer.from_pretrained("plguillou/t5-base-fr-sum-cnndm")
model = T5ForConditionalGeneration.from_pretrained("plguillou/t5-base-fr-sum-cnndm")

To summarize an ARTICLE, just modify the string like this : "summarize: ARTICLE".

Training data

The base model I used is JDBN/t5-base-fr-qg-fquad (it can perform question generation, question answering and answer extraction).

I used the "t5-base" model from the transformers library to translate in french the CNN / Daily Mail summarization dataset.

Downloads last month
843
Inference Examples
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated) instead.

Model tree for plguillou/t5-base-fr-sum-cnndm

Finetunes
5 models

Dataset used to train plguillou/t5-base-fr-sum-cnndm

Spaces using plguillou/t5-base-fr-sum-cnndm 5