fabiocigaina/deberta-v3-base-nli-fever-pietrolesci

DeBERTa-v3-base NLI (FEVER-style)

This model is a Natural Language Inference (NLI) classifier fine-tuned on a FEVER-style dataset. Given a claim and an evidence sentence, the model predicts whether the evidence entails, is neutral to, or contradicts the claim.

Links

• Demo: https://huggingface.co/spaces/fabiocigaina/nli-fever-demo
• Code: https://github.com/FabioCigaina/FEVER-Claim-Verification

Task

• Type: Text Classification (Natural Language Inference)
• Input: (claim, evidence)
• Output labels:
  • 0: entailment
  • 1: neutral
  • 2: contradiction

Base Model

microsoft/deberta-v3-base

Dataset

• pietrolesci/nli_fever
• Preprocessing:
  • Removed samples with empty evidence fields

Training

• Epochs: up to 10
• Early stopping:
  • Metric: F1 macro
  • Best checkpoint reached after 5 effective epochs

Evaluation Metrics

Metric	Score
Accuracy	0.76
F1 (macro)	0.75