Diffusion Reward on Language Environment

This repository implements the Diffusion Reward pipeline on the Google Language Table environment

Set the Python Path in root directory

cd diffusion_language
export PYTHONPATH=$PYTHONPATH:$(pwd)

Set up Diffusion_Reward environment:

cd diffusion_reward
export PYTHONPATH=${PWD}:$PYTHONPATH

Create a virtual environment.

conda env create --name diffusion_language -f conda_env.yml 
conda activate diffusion_reward
pip install -e .

Install extra dependencies.

• Install PyTorch.

pip3 install torch==1.12.1+cu116 torchvision==0.13.1+cu116 --extra-index-url https://download.pytorch.org/whl/cu116

Hack to fix failed to initialize OpenGL error

unset LD_PRELOAD

Set up Language_Table environment:

cd language_table
pip install -r ./requirements_static.txt
export PYTHONPATH=${PWD}:$PYTHONPATH

Training

Run the RL training loop for the blocktoblock64 task, which used VQ-VAE to train on 64x64 frames of 'move the green star to the blue cube'

bash scripts/run/rl/drqv2_langauge_diffusion_reward.sh blocktoblock64