---
library_name: transformers
datasets:
- facebook/xnli
metrics:
- accuracy
base_model:
- FacebookAI/xlm-roberta-large
license: mit
tags:
- xlm-roberta
- finetuning
- xnli
- mnli
---

# XLM-RoBERTa Large finetuned on XNLI dataset

<!-- Provide a quick summary of what the model is/does. -->

## Model Details

- **Developed by:** Adrien J.
- **Model type:** XLM-RoBERTa
- **Languages (NLP):** Multilingual
- **Finetuned from model:** [FacebookAI/xlm-roberta-large](https://huggingface.co/FacebookAI/xlm-roberta-large)

## How to Get Started

This model is ready-to-use for text classification.

```py
import pandas as pd
from transformers import pipeline

# Load the classification pipeline
classifier = pipeline("text-classification", "ajayat/xlm-roberta-large-xnli")
classifier.model.config.id2label = {
    0: "entailment", 
    1: "neutral", 
    2: "contradiction"
}
# Example premise and hypothesis
premise = "A soccer game with multiple males playing."
hypothesis = "Some men are playing a sport."

# Provide input as a dictionary with text and text_pair keys
result = classifier({'text': premise, 'text_pair': hypothesis}, top_k=None)
pd.DataFrame(result)
```

|   | label	        | score    |
|:-:|:-------------:|:--------:|
| 0 | entailment	| 0.996513 |
| 1 | neutral	    | 0.003228 |
| 2 | contradiction	| 0.000260 |

## Dataset

The [XNLI dataset](https://huggingface.co/datasets/facebook/xnli) (Cross-lingual Natural Language Inference) is a benchmark dataset 
created by Facebook AI for evaluating cross-lingual understanding. 
It extends the MultiNLI corpus by translating 7,500 human-annotated English sentence pairs (premise and hypothesis) into 14 languages.

Each pair is labeled as `entailment`, `contradiction`, or `neutral`.

## Training Hyperparameters

- `bf16` mixed precision
- Batch size per GPU: `64`
- Learning Rate: `2e-5`
- 1 Epoch

## Results

Here are the results on the XNLI test set:

| lang_abv |  ar  |  bg  |  de  |  el  |  en  |  es  |  fr  |  hi  |  ru  |  sw  |  th  |  tr  |  ur  |  vi  |  zh  | avg  |
|:--------:|:----:|:----:|:----:|:----:|:----:|:----:|:----:|:----:|:----:|:----:|:----:|:----:|:----:|:----:|:----:|:----:|
| accuracy | 0.82 | 0.85 | 0.85 | 0.85 | 0.89 | 0.85 | 0.84 | 0.81 | 0.83 | 0.77 | 0.81 | 0.82 | 0.77 | 0.83 | 0.83 | 0.84 |


## Environmental Impact

<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->

Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).

- **Hardware Type:** 4x GPUs NVIDIA A100 SXM4 80GB
- **Hours used:** 7 hours
- **Compute Region:** France