| license: mit | |
| task_categories: | |
| - multiple-choice | |
| - question-answering | |
| language: | |
| - en | |
| tags: | |
| - stem | |
| - mcq-solver | |
| - flan-t5 | |
| - minilm | |
| # Smart MCQ Solver Challenge | |
| This repository contains the machine learning pipeline and model solution for the **Smart MCQ Solver Challenge**. | |
| ## Models Included | |
| 1. **Model 1 (Scratch)**: GBDT trained on TF-IDF lexical features | |
| 2. **Model 2 (Pretrained)**: `google/flan-t5-large` (770M) zero-shot token probability ranker | |
| 3. **Model 3 (Ensemble)**: Soft-voting blend of `google/flan-t5-large` and `sentence-transformers/all-MiniLM-L6-v2` (22M) | |
| ## Usage | |
| Run `app.py` or `smart_mcq_solver_complete.ipynb` to execute predictions. | |