logicalqubit/deberta-v3-large-news-classifier

DeBERTa v3 Large News Classifier

This model is a fine-tuned version of microsoft/deberta-v3-large for multi-class classification of news headlines. The model classifies headlines into one of 7 categories:

• World
• Business
• Technology
• Entertainment
• Sports
• Science
• Health

---

Training Dataset

This model was trained on the logicalqubit/news_133k dataset,
which contains 133,000 labeled news headlines across the 7 categories mentioned above.

---

Training Hyperparameters:

• learning_rate: 6e-6

• train_batch_size: 8

• eval_batch_size: 8

• warmup_steps: 50

• num_epochs: 2

• report_to: wandb

---

Sample Code

---

Sample Output

---

Metrics

• While zero-shot models and fine-tuned text classification model differ fundamentally in design and purpose, this comparison is presented for reference. It provides a practical benchmark, even if the comparison is entirely unfair.

---

Wandb

---