âī¸ Quantum LLM Training
Quantum Student Learning Pipeline
Train quantum-compressed models using knowledge distillation from pretrained classical teacher models.
Quantum Training Pipeline
1
Teacher Model
Load Pretrained Classical Model
2
Quantum Student
Initialize Quantum Circuit
3
Distillation
Knowledge Transfer
4
Compression
Quantum Compression
5
Training
Quantum-Enhanced Training
âī¸ Quantum Training Configuration
Recommended: Medical pretrained model (checkpoint-5000) available on Teraq Backend instance.
This model is trained on medical data and is accessible from the quantum training instance.
Located at:
/home/ec2-user/Training_Data/models/tinyllama-1b-medical-phase1-48vcpu/checkpoint-5000
Hold Ctrl/Cmd to select multiple
Higher temperature = softer teacher predictions
Balance between teacher and student loss (0.7 = 70% teacher, 30% student)
đ TensorBoard Status
Checking TensorBoard status...
Loading TensorBoard information...
đ Training Logs Terminal
Checking for running training...
Loading training logs...
âī¸ Quantum Distillation
Compress models using quantum-inspired algorithms for 2x-5x size reduction while maintaining performance. The quantum student learns from the classical teacher through knowledge distillation.
đŦ 29-Qubit Quantum Circuits
Advanced quantum ansatz layers with 29 qubits for enhanced model training and optimization. Uses MPS (Matrix Product State) backend for efficient simulation.
đ Hybrid Classical-Quantum Training
Seamlessly integrate quantum and classical training methods. Start with a pretrained classical teacher model and distill knowledge into a quantum-compressed student model.