DistilBERT for AG News Classification — Full Fine-Tune
This repository contains a DistilBERT model fine-tuned on the AG News dataset.
Results
| Model | Test Accuracy | Macro F1 |
|---|---|---|
| Full Fine-Tune | 0.9426 | 0.9427 |
Confusion Matrix (Test)
World Sports Busines Sci/Tec
World 1797 13 42 48
Sports 10 1871 13 6
Busines 39 3 1692 166
Sci/Tec 26 8 62 1804
Usage
from transformers import pipeline
clf = pipeline("text-classification", model="starkdv123/agnews-distilbert-ft")
clf(["Markets tumble as central bank raises rates."])
Training (summary)
- Base:
distilbert-base-uncased - Epochs: 2, LR: 2e-5, batch: 16, max_len: 256, weight_decay: 0.01, fp16
Author
Karan D Vasa — https://huggingface.co/starkdv123
- Downloads last month
- 8