Islamamro/emotion6-aurora-islamamro

Emotion6 - Aurora - islamamro

Text classifier fine-tuned from distilbert-base-uncased on dair-ai/emotion (6 classes).

Built, trained and published end-to-end via the Aurora Research Portal by islamamro. Trained on CPU on a 1200-example demo subset.

• Base model: distilbert-base-uncased (~67M parameters)
• Held-out accuracy: 0.7867
• Held-out macro-F1: 0.5426

Honest scope

This is a demonstration of the Aurora build -> train -> publish pipeline on a small subset, not a production model. Model size here is set by the base architecture, which is varied deliberately across the Aurora showcase to span a range of parameter counts. For production, fine-tune on the full dataset.