🔬 Quantum-Enhanced Transformer Classification with LLM Science Dataset

This project explores various model strategies for classifying science multiple-choice questions using a high-quality dataset. It combines classical NLP methods (BERT) with quantum machine learning (QML) techniques to study their comparative performance.

---

📁 Project Structure

• bert_LLM.ipynb: Pretrained BERT fine-tuned on the full dataset.
• quantum_hybrid_custom_LLM.ipynb: SimpleTransformer with quantum kernel features appended to classical features.
• quantum_kernel_LLM.ipynb: SimpleTransformer using only quantum kernel features (100 trained, 1000 used for classification). using full training data.
• quantum_hybrid_custom_LLM.ipynb: Hybrid model combining TF-IDF features + quantum kernel.
• quantum_kernel_LLM.ipynb: Pure quantum-kernel-enhanced transformer model trained on limited data.
• bert_LLM_100_samples.ipynb: Pretrained BERT fine-tuned on the 100 samples for comparison with quantum_kernel_LLM.ipynb

---

📊 Results Summary

✅ bert_LLM.ipynb

• Model: Pretrained BERT
• Samples Used: Full dataset
• Test Accuracy: 1.00
• Classification Report:

              precision    recall  f1-score   support

           A       1.00      1.00      1.00       426
           B       1.00      1.00      1.00       363
           C       1.00      1.00      1.00       350
           D       1.00      1.00      1.00       322
           E       1.00      1.00      1.00       306

⚛️ quantum_hybrid_custom_LLM.ipynb

• Model: SimpleTransformer (TF-IDF + Quantum Kernel as input) as additional features
• Samples Used: Full dataset
• Test Accuracy: 0.9875
• Classification Report:

              precision    recall  f1-score   support

           A       0.99      0.98      0.99       426
           B       0.99      0.98      0.98       363
           C       0.99      0.99      0.99       350
           D       0.98      0.99      0.98       322
           E       1.00      0.99      0.99       306

🧠 quantum_kernel_LLM.ipynb

• Model: SimpleTransformer (trained only on quantum kernel features)
• Quantum Kernel Training: 100 samples
• Transformer Training: 1000 samples
• Test Accuracy: 0.9500
• Classification Report:

              precision    recall  f1-score   support

           A       0.96      0.90      0.93        50
           B       0.88      0.98      0.92        44
           C       0.97      0.97      0.97        30
           D       0.98      0.95      0.96        42
           E       1.00      0.97      0.99        34

---

🚀 Highlights

• 🤖 Pretrained BERT achieves perfect performance on this task when trained on the full dataset.

• ⚛️ Quantum-hybrid models offer nearly equivalent results while using simpler architectures.

• 🧠 Quantum-only models demonstrate that competitive performance is possible with significantly fewer training samples.

• ✅ Quantum-enhanced approaches achieve strong performance, even with limited data.

• ⚛️ Hybrid models slightly underperform pure BERT but generalize well.

• 🔬 Kernel-only models can match classical models with fewer samples, opening the door to quantum-efficient learning.

---

📦 Requirements

pip install qiskit qiskit-machine-learning transformers datasets scikit-learn torch

---

📘 Citation

Dataset: lizhecheng/llm-science-dataset