Requirements: Python 3.12.3, CUDA 12.4 (for GPU), Linux (Ubuntu 20.04+ recommended), conda & pip.

Steps:

1. Create Conda env: conda create -n mdpr_project python=3.12.3 && conda activate mdpr_project

2. Install deps: pip install -r requirements.txt

3. Verify: python -c "import torch; print(torch.__version__, torch.cuda.is_available())" (expect 2.5.1+cu124 True)

Run:

1. Prepare dataset in ./data (e.g., data/train, data/val).

2. Run: python main_maple.py --dataset cifar100 --batch_size 128 --epochs 20 --lr 0.001

3. Outputs: Model weights in ./checkpoints/cifar100/ (e.g., best_best.pth.tar), logs and test results in ./checkpoints/cifar100/ (e.g., mdpr_clip_maple_log.txt, mdpr_clip_maple_best.txt).