README.md

QSPLIT: Quantum Split Learning Testing Toolkit

QSPLIT is a GUI-based testing framework designed for the efficient validation of quantum split learning architectures by automatically generating dummy code to complete user-provided QNN component subsets. It supports parallel split learning across diverse architectural combinations, allowing developers to monitor real-time logs and identify trainable configurations through an intuitive interface.

---

🛠 Tool Components

A. Part Selection

Define which components of the QNN (SE, PQC, MEA) are provided by the user and which are generated.

• Target Code: User-uploaded core logic.
• Dummy Code: Components automatically generated to ensure structural compatibility.

B. Target Code Hyperparameters

• Quantum Device: Configures the number of qubits, batch size, and circuit depth.
• Training: Defines epochs, optimizer, and learning rate for the split learning process.

C. Dummy Code Generation

• Automatically generates variations of dummy codes (e.g., PQC layers or MEA observables).
• The Dummy List allows users to inspect the internal gate operations of each generated code.

D. Split Learning Execution

• Powered by the TorchQuantum framework.
• Executes parallel training across multiple target-dummy combinations with real-time log streaming.

E. Result Visualization & Export

• Displays performance metrics (Accuracy/Loss) for all combinations.
• Validated components can be exported as executable .py files for seamless integration into production.

---

🚀 HOW TO USE

1. Getting Started

💻 Local Setup

Prerequisites:

• Python ≥ 3.12 | Flutter 3.32.8
• PyTorch 2.8.0 | TorchQuantum 0.1.8

Step 1: Environment Setup Clone the repository and install dependencies. We recommend using Linux or WSL.

Option A: Using pip (requirements.txt)

If you prefer a standard Python environment:

pip install --upgrade pip
pip install -r requirements.txt

Option B: Using Conda (environment.yml)

If you prefer Conda, we recommend configuring the libmamba solver for faster dependency resolution:

conda update -n base conda
conda install -n base conda-libmamba-solver
conda config --set solver libmamba
conda env create -f environment.yml

Step 2. Dataset Setup Before running the backend, you must download and place the dataset.

• Download Link: Dataset
• Path: Place the downloaded images into the following directory: Backend/dataset/medmnist/AbdomenCT

Step 3. Run Application

# run backend
cd Backend
uvicorn app.main:app --host 0.0.0.0 --port 8000

# run frontend
cd Frontend
flutter run -d chrome

2. Configure Components

• Upload Target Code: In Part Selection, click Upload to provide any subset of SE / PQC / MEA as target code.
• Component Selection: Select each component as Target Code or Dummy Code (radio buttons; mutually exclusive).
• Parameter Input: In Target Code Hyperparameter
  • Quantum Device: number of qubits, batch size, circuit depths
  • Training: epochs, optimizer, learning rate
  • Dataset: choose from datasets uploaded to QSPLIT (e.g., MedNIST)

3. Generate Dummy Code

1. Configure parameters in Target Code Hyperparameter
2. Click Generate to create dummy code
3. Inspect in Dummy Code Generation → Dummy List
   • Example: PQC = RY/RZ/CNOT gate stack, MEA = Z observable

4. Execute Split Learning

• In Dummy Code Generation, click Run to start split learning across all combinations
• Training uses TorchQuantum; per-epoch loss/accuracy are streamed in real time to the GUI

5. Compare Results

• Results section lists accuracy and training time per combination
• Easily identify stable or high-performing combinations

6. Export Code

• In Results, select a dummy code and click Export
• Outputs an executable .py file compatible with TorchQuantum
• Enables reuse, extension, and integration into follow-up experiments or deployment

Supplementary Examples

Target Code: MEA, Dummy Code: SE, PQC

QSPLIT_MEA.mp4

Software Version

v1.0.0