RayyanAlam123/roberta-large-bias-detector

KhabarCheck - RoBERTa-Large Bias Model

Fine-tuned roberta-large for political bias classification in news articles for KhabarCheck.

Model Details

Base model	roberta-large
Task	3-class text classification
Classes	0 = Left 🔵, 1 = Center ⚖️, 2 = Right 🔴
Accuracy	85.43%
F1 (weighted)	85.39%
Max tokens	512

Usage

from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

model_id = "RayyanAlam123/roberta-large-bias-detector"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForSequenceClassification.from_pretrained(model_id)
model.eval()

text = "The president signed a new executive order on climate policy."
inputs = tokenizer(text, return_tensors="pt", max_length=512, truncation=True)
with torch.no_grad():
    logits = model(**inputs).logits
probs = torch.softmax(logits, dim=-1).squeeze()
labels = ["Left", "Center", "Right"]
print(f"Prediction: {labels[probs.argmax()]}  ({probs.max()*100:.1f}% confidence)")

Training Details

Hyperparameter	Value
Learning rate	1e-5
LR scheduler	cosine
Effective batch size	32 (4 × 8 grad accum)
Warmup	8%
Mixed precision	bf16
Early stopping patience	3

Dataset

10,020 balanced news articles — 3,340 each for Left, Center, Right.

Live Demo

Hugging Face Space