ethnmcl/roberta-articulation-classifier

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

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