Compute Resources
Overview of EC2 instances, quantum computing backends, and infrastructure resources
EC2 Instances
Teraq Backend
Running t3.2xlarge
Hostname
ec2-13-223-206-81.compute-1.amazonaws.com
IP Address
13.223.206.81
vCPUs
< 40 vCPU
SSH Access
ssh -i Teraq.pem ec2-user@ec2-13-223-206-81.compute-1.amazonaws.com
API Server
Port 8000
Purpose
Classical & Quantum Training
Services:
QTinyLlama
QRoBERTa
QVIT
Training APIs
PostgreSQL
48vCPU Training Instance
Running 48 vCPU
Hostname
ec2-204-236-243-64.compute-1.amazonaws.com
IP Address
204.236.243.64
vCPUs
48 vCPU
Status
Training Active
Current Process
train_tinyllama_48vcpu.py
Note
Do not run additional training
Usage:
Large-Scale Training
TinyLlama-1.1B
Medical Domain
Summit Backend Models
Running EC2
Hostname
ec2-34-204-191-199.compute-1.amazonaws.com
IP Address
34.204.191.199
SSH Access
ssh -i Teraq.pem ec2-user@ec2-34-204-191-199.compute-1.amazonaws.com
Storage
100GB (56GB used, 45GB available)
Purpose
Model Inference & Training
Services:
TinyLlama API (Port 8100)
QRoBERTa API (Port 8101)
Model Training
Model Storage
Forms API Server
Running EC2
Hostname
ec2-18-205-155-235.compute-1.amazonaws.com
IP Address
18.205.155.235
API Port
8600
API Endpoint
http://18.205.155.235:8600
Purpose
Forms Processing & Batch Management
Services:
FastAPI Forms Service
PostgreSQL
Batch Review
Training Logs
Teraq Backend (Alternative)
Running EC2
Hostname
ec2-54-242-218-168.compute-1.amazonaws.com
IP Address
54.242.218.168
Purpose
Reference Architecture
Services:
Quantum Pipelines
Training Templates
PostgreSQL + pgvector
Quantum Computing Resources
Qiskit Quantum Simulators
QASM Simulator
Simulator
Type: Quantum Assembly Language Simulator
Usage: General-purpose quantum circuit simulation
Qubits: Up to 30+ qubits (simulated)
Access:
Applications: QTinyLlama quantum circuits, QRoBERTa MPS layers, general VQC training
Usage: General-purpose quantum circuit simulation
Qubits: Up to 30+ qubits (simulated)
Access:
from qiskit import Aer; backend = Aer.get_backend('qasm_simulator')Applications: QTinyLlama quantum circuits, QRoBERTa MPS layers, general VQC training
Statevector Simulator
Simulator
Type: Full statevector simulation
Usage: Exact quantum state evolution (no measurement noise)
Qubits: Up to 25-30 qubits (memory-limited)
Access:
Applications: Quantum gradient computation, exact expectation values, debugging quantum circuits
Usage: Exact quantum state evolution (no measurement noise)
Qubits: Up to 25-30 qubits (memory-limited)
Access:
backend = Aer.get_backend('statevector_simulator')Applications: Quantum gradient computation, exact expectation values, debugging quantum circuits
Matrix Product State (MPS) Simulator
Simulator
Type: Tensor network simulator (O(n) complexity)
Usage: Efficient simulation of low-entanglement quantum states
Qubits: Up to 100+ qubits (bond dimension dependent)
Bond Dimension: Configurable (default: 32-64)
Access:
Applications: QRoBERTa quantum layers, QVIT quantum attention, large-scale quantum ML
Usage: Efficient simulation of low-entanglement quantum states
Qubits: Up to 100+ qubits (bond dimension dependent)
Bond Dimension: Configurable (default: 32-64)
Access:
backend = Aer.get_backend('matrix_product_state')Applications: QRoBERTa quantum layers, QVIT quantum attention, large-scale quantum ML
Real Quantum Hardware (Available)
IBM Quantum Hardware
Hardware
Provider: IBM Quantum Network
Available Systems: ibm_brisbane (127 qubits), ibm_kyoto (127 qubits), ibm_osaka (127 qubits)
Access: Requires IBM Quantum account and API token
Setup:
Queue Time: Minutes to hours (depending on system load)
Cost: $1-10 per job (varies by system)
Applications: Production quantum ML, real quantum advantage validation
Available Systems: ibm_brisbane (127 qubits), ibm_kyoto (127 qubits), ibm_osaka (127 qubits)
Access: Requires IBM Quantum account and API token
Setup:
from qiskit_ibm_runtime import QiskitRuntimeService; service = QiskitRuntimeService()Queue Time: Minutes to hours (depending on system load)
Cost: $1-10 per job (varies by system)
Applications: Production quantum ML, real quantum advantage validation
D-Wave Quantum Annealing
Hardware
Provider: D-Wave Systems
Qubits: 5000+ qubits (Advantage systems)
Type: Quantum annealing (optimization-focused)
Access:
Best For: QUBO/QUBOOST optimization, combinatorial optimization
Speedup: 100x faster on 100k-10M file processing
Applications: Q_Physical_AI models, large-scale optimization problems
Qubits: 5000+ qubits (Advantage systems)
Type: Quantum annealing (optimization-focused)
Access:
pip install dwave-ocean-sdkBest For: QUBO/QUBOOST optimization, combinatorial optimization
Speedup: 100x faster on 100k-10M file processing
Applications: Q_Physical_AI models, large-scale optimization problems
Quantum Backend Selection Guide
| Backend | Best For | Qubits | Speed |
|---|---|---|---|
| QASM Simulator | General circuits, testing | ~30 | Medium |
| Statevector | Exact gradients, debugging | ~25 | Fast (small) |
| MPS Simulator | QRoBERTa, QVIT, large circuits | 100+ | Very Fast |
| IBM Quantum | Real hardware, production | 127-433 | Queue-dependent |
| D-Wave | QUBO optimization | 5000+ | Very Fast |
Training Infrastructure
Quantum Training
QTinyLlama, QRoBERTa, and QVIT training pipelines run on Teraq Backend instance (ec2-13-223-206-81) using Qiskit quantum simulators for circuit execution.
Classical Training
Large-scale classical training (TinyLlama-1.1B) runs on 48vCPU instance (ec2-204-236-243-64) for faster training throughput.
Data Storage
PostgreSQL databases on multiple instances store training data, batch information, and model metadata. pgvector extension used for model embeddings.