question-complexity-classifier

🤖 Fine-tuned DistilBERT model for classifying question complexity (Simple vs Complex)

Model Details

Model Description

• Architecture: DistilBERT base uncased
• Fine-tuned on: Question Complexity Classification Dataset
• Language: English
• License: Apache 2.0
• Max Sequence Length: 128 tokens

Uses

from transformers import pipeline

classifier = pipeline(
    "text-classification",
    model="grahamaco/question-complexity-classifier",
    tokenizer="grahamaco/question-complexity-classifier",
    truncation=True,
    max_length=128  # Matches training config
)

result = classifier("Explain quantum computing in simple terms")
# Output example: {'label': 'COMPLEX', 'score': 0.97}

Training Details

• Epochs: 5
• Batch Size: 32 (global)
• Learning Rate: 2e-5
• Train/Val/Test Split: 80/10/10 (stratified)
• Early Stopping: Patience of 2 epochs

Evaluation Results

Metric	Value
Accuracy	0.92
F1 Score	0.91

Performance

Metric	Value
Inference Latency	15.2ms (CPU)
Throughput	68.4 samples/sec (GPU)

Ethical Considerations

This model is intended for educational content classification only. Developers should:

• Regularly audit performance across different question types
• Monitor for unintended bias in complexity assessments
• Provide human-review mechanisms for high-stakes classifications
• Validate classifications against original context when used with RAG systems