| language: en | |
| license: mit | |
| tags: | |
| - text-classification | |
| - roberta | |
| - education | |
| - student-articulation | |
| # RoBERTa Student Articulation Classifier | |
| Fine-tuned `roberta-base` for classifying student coding check-ins into 4 articulation levels. | |
| ## Labels | |
| | ID | Label | Description | | |
| |----|-------|-------------| | |
| | 0 | Minimal | Vague, no technical detail | | |
| | 1 | Basic | Names a tool or action, no implementation detail | | |
| | 2 | Developing | Specific and technical but incomplete | | |
| | 3 | Proficient | Full technical ownership — implementation, reasoning, testing | | |
| ## Usage | |
| ```python | |
| from transformers import pipeline | |
| classifier = pipeline( | |
| "text-classification", | |
| model="your-username/roberta-articulation-classifier" | |
| ) | |
| result = classifier("I implemented JWT auth with refresh token rotation and wrote integration tests.") | |
| print(result) | |
| # [{'label': 'Proficient', 'score': 0.92}] | |
| ``` | |
| ## Training | |
| - Base model: `roberta-base` | |
| - Dataset: Synthetic student check-in dataset (~580 examples) | |
| - Regularization: Label smoothing, dropout 0.2, weight decay 0.1 | |
| - Epochs: 7 (early stopping) | |
| - Test accuracy: 93.75% | |
| - Proficient F1: 1.00 | |