# Set the CUDA devices to use
export CUDA_VISIBLE_DEVICES=0,1,2,3;

# Launch the distributed training
python -m torch.distributed.launch \
    --nproc_per_node=4 \
    --use_env \
    --master_port=29501 \
    3_train_fill50k_rescale.py \
        --batch-size 32 \
        --total_train_steps 50000 \
        --teacher-model vit_large \
        --target_model vit_base \
        --model models_proteus_dinov2 \
        --patch_size 14 \
        --cldm_learning_rate 1e-4 \
        --lambda_token 1.0 \
        --lambda_fea 1.0 \
        --lambda_patch 1.0 \
        --finetune "/home/t2vg-a100-G4-1/projects/qiyp/1_feature_extractor/1_feature_extractor/checkpoint0160.pth" \
        --log_dir '/home/t2vg-a100-G4-1/projects/qiyp/3_ControlNet/log/' \
        --resume_path '/home/t2vg-a100-G4-1/projects/qiyp/3_ControlNet/models/control_sd15_fill50k.ckpt' \
        --image_floder '/home/t2vg-a100-G4-1/projects/dataset/train2017' \
        --txt_path '/home/t2vg-a100-G4-1/projects/dataset/annotations/captions_train2017.json' \
        --log_every 25 \
        --ckpt_every 2500 \
        --image_every 100 \
        --cldm_yaml './models/cldm_fill50k.yaml' \
        --exp_dir './exp_fill50k_rescale' \
        --controlnet_dir '/home/t2vg-a100-G4-1/projects/qiyp/3_ControlNet/exp_fill50k_rescale/checkpoints/0002500.pt' \