ethnmcl's picture
Upload README.md with huggingface_hub
ccab90d verified
|
Raw
History Blame Contribute Delete
1.15 kB
---
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