HW 2 - DistilBERT fine-tuned
Task
Fine-tuned DistilBERT for text classification on a classmate's HW1 dataset.
- Dataset:
ddecosmo/hw_text_dataset - Text column:
Text - Label column:
label(classes: ['asu', 'bucknell', 'cmu', 'duq', 'ucsd']) - Train/Eval split: 80/20 (stratified if available)
Training
- Base model:
distilbert-base-uncased - Epochs: 3, LR=5e-5, WD=0.01, warmup=10%
- Batch size: 16
- Best model by: F1 (macro)
Results (Test)
- Accuracy: 0.4000
- F1 (macro): 0.1231
- Precision (macro): nan
- Recall (macro): nan
Notes & Limitations
- Small student dataset; results may vary with seeds.
- Labels mapped as: {'asu': 0, 'bucknell': 1, 'cmu': 2, 'duq': 3, 'ucsd': 4}
AI Tool Disclosure
This notebook used ChatGPT for scaffolding code and documentation. All dataset selection, training, evaluation, and uploads were performed by the student.
- Downloads last month
- 10