---
license: mit
---

Fine-tuned Twitter XLM Roberta Base Model from cardiffnlp: [model](https://huggingface.co/cardiffnlp/twitter-xlm-roberta-base) on French part of XNLI dataset \
using xnli finetuning script. 

Hyperparameters: 
 - Learning Rate: 5e-5
 - Epochs : 2.0
 - Batch size: 32
 - Max sequence length 128

Train Runtime:
 - 6462 s  --> 107.7 min on one NVIDIA GeForce RTX 3090

Results French XNLI 
- Eval set : 77.75
- Test set: 77.72