model (deeplearning): Camembert baseline

[ezalos]

📖 Describe what you want

Make a deeplearning baseline with Hugging Face and Pytorch.

Relevant tutorials to follow:

• Most relevant (using camembert): https://ledatascientist.com/analyse-de-sentiments-avec-camembert/
• https://www.kaggle.com/chayan8/sentiment-analysis-using-bert-pytorch
• https://skimai.com/fine-tuning-bert-for-sentiment-analysis/

Relevant documentation:

• HuggingFace: https://huggingface.co/course/chapter1/1
• (Not yet necessary, but still cool to read) PyTorch: https://pytorch.org/tutorials/beginner/basics/intro.html

✔️ Definition of done

1. Being able to train the model then save it's weights poetry run python -n src models --model camembert --output-weights models/xxxxx --train-split xxx.csv
2. Save the weights in remote s3 dvc add models/xxxxx git add && git commit dvc push -r s3-remote
3. Being able to load model weights and predict on a given dataset
4. Create a unit test which train and predict from a csv of 100 examples which is push to github (in src/tests/test_dataset.csv)
5. If necessary update the .github/workflows/cicd.yaml and the .42AI/pre-commit.git to pass the new unit test

[madvid]

I am starting the first part, I will used the IMDB if test with data is necessary

[madvid]

Interesting library developped by hugginface https://github.com/huggingface/datasets
In particular if we want to follow the first ressource gave by ezalos