RT3

step1: Download the pre-compressed model

• BaiDu NetDisk link: https://pan.baidu.com/s/1OIV08F4tUqyFD1DGfbkJ5g
• Extraction code：1dtv

Or

• Google Drive link: https://drive.google.com/drive/folders/1mOYzWXVU6SWt34v0N1KOl62_vv04R6-n?usp=sharing

step2: Environment

conda env create -f pytorch_BCM_pruning.yml

• python=3.6.9
• pytorch=1.3.0
• tensorboard=1.14.0
• torchtext=0.4.0
• torchvision=0.4.1
• ......

step3: start from rl_controller.py

• Modify the rl_input.py "timing_constraint":115#115 for high, 104 for middle,94 for low for different timing constraints
• run: CUDA_VISIBLE_DEVICES=XXX nohup python -u d_rl_controller.py > result