1. Run updates

sudo apt-get update -y -qq sudo apt-get upgrade -y -qq sudo apt-get install -y -qq golang neofetch zsh byobu sudo apt-get install -y -qq software-properties-common sudo add-apt-repository -y ppa:deadsnakes/ppa sudo apt-get install -y -qq python3.12-full python3.12-dev sh -c "$(curl -fsSL https://raw.githubusercontent.com/ohmyzsh/ohmyzsh/master/tools/install.sh)" "" --unattended sudo chsh $USER -s /usr/bin/zsh

2. Git clone

3. Config the venv

python3.12 -m venv ~/Landcover/.venv . ~/Landcover/.venv/bin/activate pip install -U pip pip install -U wheel pip install -U "jax[tpu]" -f https://storage.googleapis.com/jax-releases/libtpu_releases.html pip install -U equinox pip install -U optax pip install -U grain pip install -U tqdm pip install -U typing pip install -U tensorflow_datasets pip install -U PyYAML pip install -U pytest pytest-benchmark pip install -U tensorboardX pip install -U matplotlib pip install -U tensorflow tensorboard-plugin-profile

For profiling, if you dont want tensorboard profiling for jax, just skip

4. Mount the dataset

login : gcloud auth application-default login

install google fuse

export GCSFUSE_REPO=gcsfuse-`lsb_release -c -s`
echo "deb https://packages.cloud.google.com/apt $GCSFUSE_REPO main" | sudo tee /etc/apt/sources.list.d/gcsfuse.list
curl https://packages.cloud.google.com/apt/doc/apt-key.gpg | sudo apt-key add -
sudo apt-get update
sudo apt-get install fuse gcsfuse
gcsfuse -v

sudo mkdir "$HOME/dataset" sudo chmod 777 "$HOME/dataset" gcsfuse --implicit-dirs --implicit-dirs --file-mode 777 --dir-mode 777 --uid=0 --gid=0 s2glc_array_records "$HOME/dataset"

5. Mount the logs dir

TODO :

• Clean up the metrics and tensorboard profiling / logs
• Add visualization to the training to track the progress between epochs
• Add more unit tests to ensure every module is working as expected especially the sharding between the TPU cores
• If everything is working as expected move on to DDPM and try to implement it in JAX