Preparation

• conda env create -f environment.yml
• conda activate qbridge
• pip install -r requirements-ml.txt
• export TF_USE_LEGACY_KERAS=True

QOIN

The code in the QOIN folder is copied and revised from the paper Mitigating Noise in Quantum Software Testing Using Machine Learning (link).

• Run python QOIN/DataGeneration.py to generate baseline circuits.
• Run python QOIN/MLPTraining.py to pretrain QOIN’s baseline models.
• Run python QOIN/BaselineTuner.py to generate CUTs and finetune QOIN’s tuning models.
• Run python QOIN/EvaluationRQ1.py to generate testing circuits.
• Run python QOIN/CalculatingRQ1.py to obtain QOIN’s results for RQ1.
• Run python QOIN/EvaluationRQ3.py and python QOIN/CalculatingRQ3.py to obtain QOIN’s results for RQ3.

QLEAR

The code in the QLEAR folder is copied and revised from the paper A Machine Learning-Based Error Mitigation Approach for Reliable Software Development on IBM’s Quantum Computers (link).

• Run python QLEAR/DataGeneration_QLEAR_Pretrain.py to generate baseline circuits.
• Run python QLEAR/QLEAR_Pretrain_MLP_ByBackend.py to pretrain QLEAR’s baseline models.
• Run python QLEAR/QLEAR_Finetune_Data_ByBackendFamily.py to generate CUTs circuits.
• Run python QLEAR/QLEAR_Finetune_MLP_BySeedBackendFamily.py to finetune QLEAR’s tuning models.
• Run python QLEAR/QLEAR_Test_Data_ByBackendFamily.py to generate testing circuits.
• Run python QLEAR/QLEAR_Evaluate_Hellinger_BySeedBackendFamily.py to obtain QLEAR’s results for RQ1.

QRAFT

The code in the QRAFT folder is copied and revised from the paper Qraft: reverse your quantum circuit and know the correct program output (link).

• Run python QRAFT/QraftFeatureGeneration.py to generate baseline circuits.
• Run python QRAFT/QraftEDTPretrain.m to pretrain QRAFT’s baseline models.
• Run python QRAFT/QraftFamilyTuneDataGeneration.py to generate CUTs circuits.
• Run python QRAFT/QraftEDTFinetune.m to finetune QRAFT’s tuning models.
• Run python QRAFT/QraftTestDataGeneration.py to generate testing circuits.
• Run python QRAFT/QQraftEDTEvaluateTest.m to obtain QRAFT’s results for RQ1.

data_preparation

This folder contains auxiliary scripts for data processing and preprocessing tasks.
They are used to prepare datasets before running the main experiments and training pipelines.

transformer

For Q-Bridge backend-wise model:

• Run pretrain.sh for pretraining
• Run finetune.sh or finetune_parallel.sh for finetuning
• Run test.sh for testing

For Q-Bridge general model:

• Run pretrain_full_model.sh for pretraining
• Run finetune_full_model.sh for finetuning
• Run test_full_model.sh for testing

ablation study

For Q-Bridge backend-wise model:

• Run pretrain_ablation.sh for pretraining
• Run finetune_ablation.sh for finetuning
• Run test_ablation.sh for testing

For Q-Bridge general model:

• Run pretrain_full_model.sh for pretraining
• Run finetune_full_model.sh for finetuning
• Run test_full_model.sh for testing

setting

• Remove Edge-Biased Attention: ./transformer/config.yaml -> use_edge_bias: false
• Remove FiLM: ./transformer/config.yaml -> use_film: false
• Remove backend embedding: replace 'import model' with 'import model_v2' for all the files
• Add qubit multi-hot: replace 'import circuit_dag_converter_v2' with 'import circuit_dag_converter' for all the files

larger_circuits

• Run python larger_circuits/BaselineTuner.py to generate circuits for finetuning and testing.
• Run python larger_circuits/EvaluationRQ1.py to generate datasets for finetuning and testing.
• Refer to transformer for finetuning and testing model