Vantish/bert-based-uncased-agnews4-v01

bert-base-uncased-agnews4-v01

Model description

This model is a fine-tuned version of bert-base-uncased on the AG News dataset. It classifies English news headlines or articles into four categories:

The model was trained for text classification using the transformers library from Hugging Face.

Intended uses & limitations

Intended uses

• —News topic classification
• —Benchmarking or educational purposes
• —Text categorization for NLP pipelines

Limitations

• —The model is trained on English text only.
• —It may perform poorly on informal or domain-specific text (e.g., social media).
• —The dataset contains news articles up to 2015; recent topics or events may be underrepresented.

How to use

from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

tokenizer = AutoTokenizer.from_pretrained("Vantish/bert-based-uncased-agnews4-v01")
model = AutoModelForSequenceClassification.from_pretrained("Vantish/bert-based-uncased-agnews4-v01")

text = "NASA launches a new satellite to study climate change."
inputs = tokenizer(text, return_tensors="pt")
outputs = model(**inputs)
pred = torch.argmax(outputs.logits, dim=-1).item()

label_map = {
    0: "World",
    1: "Sports",
    2: "Business",
    3: "Sci/Tech"
}
print("Predicted label:", label_map[pred])

Training procedure

• —Base model: bert-base-uncased
• —Dataset: AG News
• —Task: Text classification (4 classes)
• —Optimizer: AdamW
• —Learning rate: 2e-5
• —Batch size: 16
• —Epochs: 3
• —Loss function: CrossEntropyLoss
• —Evaluation metric: Accuracy

Hardware

Trained on a single NVIDIA GPU (e.g., T4 or V100).

Evaluation results

| Metric | Score | | F1 Score | 85.6% | | Accuracy | 85.5% |

⚠️ Replace with your actual evaluation results if available.

Citation

If you use this model, please cite:

@misc{bert_agnews_finetune,
  title={BERT Base Uncased Fine-Tuned on AG News},
  author={Vantish},
  year={2025},
  howpublished={\url{https://huggingface.co/Vantish/bert-based-uncased-agnews4-v01}},
}

License

• —Base model: Apache 2.0 License
• —Fine-tuned weights: Apache 2.0

Author

Author: [Vantish]