---
tags:
- NLI
- Natural Language Inference
- FEVER
- text-classification
language: en
task: NLI and generation of Adversarial Examples
datasets: FEVER
license: unknown
metrics:
  epoch:
  - 0
  train_loss:
  - 0.0019978578202426434
  val_loss:
  - 2.2035093307495117
  train_acc:
  - 1.0
  val_acc:
  - 0.7333915829658508
  train_f1_score:
  - 1.0
  val_f1_score:
  - 0.7333915829658508
  best_metric: 2.2035093307495117
model-index:
- name: nli-fever
  results:
  - task:
      type: nlp
      name: Multi-Lingual Natural Language Processing
    dataset:
      name: FEVER
      type: fever
    metrics:
    - type: acc
      value: '0.73'
      name: Accuracy
      verified: false
---

# NLI-FEVER Model

This model is fine-tuned for Natural Language Inference (NLI) tasks using the FEVER dataset.

## Model description

This model is based on roberta and has been fine-tuned for NLI tasks.
It classifies a given pair of premise and hypothesis into three categories: entailment, contradiction, or neutral.

## Intended uses & limitations

This model is intended for use in NLI tasks, particularly those related to fact-checking and verifying information.
It should not be used for tasks it wasn't explicitly trained for.

## Training and evaluation data

The model was trained on the FEVER (Fact Extraction and VERification) dataset.

## Training procedure

The model was trained for [0] epochs
with a final loss of 2.2035093307495117, an
accuracy of 0.7333915829658508, and
F1 score of 0.7333915829658508.

## How to use

You can use this model directly with a pipeline for text classification:

```python
from transformers import pipeline

classifier = pipeline("text-classification", model="YusuphaJuwara/nli-fever")
result = classifier("premise", "hypothesis")
print(result)
```

Or, you can use it directly:

```python
from transformers import AutoTokenizer, AutoModelForSequenceClassification

tokenizer = AutoTokenizer.from_pretrained("YusuphaJuwara/nli-fever")
model = AutoModelForSequenceClassification.from_pretrained("YusuphaJuwara/nli-fever")

inputs = tokenizer("premise", "hypothesis", return_tensors="pt")
outputs = model(**inputs)
predictions = outputs.logits.argmax(-1)
print(predictions)
```

## Saved Metrics

This model repository includes a `metrics.json` file containing detailed training metrics.
You can load these metrics using the following code:

```python
from huggingface_hub import hf_hub_download
import json

metrics_file = hf_hub_download(repo_id="YusuphaJuwara/nli-fever", filename="metrics.json")
with open(metrics_file, 'r') as f:
    metrics = json.load(f)

# Now you can access metrics like:
print("Last epoch: ", metrics['last_epoch'])
print("Final validation loss: ", metrics['val_losses'][-1])
print("Final validation accuracy: ", metrics['val_accuracies'][-1])
```

These metrics can be useful for continuing training from the last epoch or for detailed analysis of the training process.

## Training results
![Include a plot of your training metrics here](training_plot.png)

Limitations and bias
## This model may exhibit biases present in the training data. Always validate results and use the model responsibly.

## Plots
![Labels distribution plots](roberta_fever/label_distribution.png)
![loss plots](roberta_fever/loss_plot.png)
![accuracy plots](roberta_fever/accuracy_plot.png)
![f1 score plots](roberta_fever/f1_score_plot.png)
![confusion matrix plots](roberta_fever/confusion_matrix.png)
![precision recall curve plots](roberta_fever/precision_recall_curve.png)
![roc curve plots](roberta_fever/roc_curve.png)