Maarij-Aqeel/BERT_news_classification

BERT News Classification (Fine-tuned on AG News)

This model is a fine-tuned version of bert-base-uncased for news topic classification on the AG News dataset.
It predicts one of four categories: World, Sports, Business, Sci/Tech.

Model Details

• Base Model: bert-base-uncased
• Fine-tuning Dataset: SetFit/ag_news
• Language: English
• Task: Text Classification (Single-label)
• Number of Labels: 4
• Training Framework: Hugging Face Transformers
• Evaluation Metric: Weighted F1-score

Training Procedure

• Epochs: 3
• Batch Size: 16
• Learning Rate: 5e-5
• Weight Decay: 0.01
• Mixed Precision: FP16 enabled
• Early Stopping: Patience of 1 epochs

Training & Evaluation Results

Epoch	Training Loss	Validation Loss	F1 Score
1	0.2005	0.1875	0.9422
2	0.1171	0.1948	0.9470
3	0.0853	0.2458	0.9457

Final Training Metrics:

• Global Step: 22500
• Training Loss: 0.1546
• Train Runtime: 2593s (~43min)
• Samples/sec: 231.39
• Steps/sec: 14.46
• Total FLOPs: 1.57e+16

Intended Use

The model can be used for:

• Classifying short news articles into World, Sports, Business, Sci/Tech.
• Benchmarking text classification pipelines with BERT.

How to Use

from transformers import AutoTokenizer, AutoModelForSequenceClassification

model_name = "Maarij-Aqeel/BERT_news_classification"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForSequenceClassification.from_pretrained(model_name)

text = "The stock market saw a significant rise today as..."
inputs = tokenizer(text, return_tensors="pt")
outputs = model(**inputs)
predicted_label = outputs.logits.argmax(dim=-1).item()
print(predicted_label)