RoBERTa-base — Smart MCQ Solver Challenge
Fine-tuned roberta-base on the Smart MCQ Solver Challenge dataset using the
AutoModelForMultipleChoice cross-encoder formulation: each of the 5 options is
scored as a separate (question, option) pair through one shared encoder, then
softmaxed into a ranking.
- Base model: roberta-base (125M params)
- Trained on: 2,000 rows, full-data deployment regime, 6 epochs
- Final training-set fit accuracy: 1.0000
- Part of: DL & GenAI Project [BSDA2001P], Smart MCQ Solver Challenge
See the project repository for the full notebook, EDA and report.
