instructions to run

Table of Contents

1. Instructions to run Locally on Docker with GPU
2. Instructions to run on Raspberry Pi (for finetuning)
3. Instructions to run on Hoffman2
4. Instructions to profile code

Instructions to run Locally on Docker with GPU

install nvidia-docker2

https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/install-guide.html

start up full gpu docker container (13.8GB)

docker build -t cs269_full -f full.dockerfile .  

docker run -ti --gpus all --shm-size=1g --ulimit memlock=-1 -e "TERM=xterm-256color" -v "$(pwd):/workspace" -v /etc/localtime:/etc/localtime:ro -v "$HOME"/.zsh_history_docker:/root/.zsh_history:z -e HIST_FILE="/root/.zsh_history" --network host cs_269 note that docker run automatically starts jupyter and tensorboard

to ssh into container

ssh -o "UserKnownHostsFile=/dev/null" -o "StrictHostKeyChecking=no" root@localhost -p 7722

to connect to jupyter notebook, copy below into browser url

localhost:8888

to connect to tensorboard

localhost:6006

Instructions to run on Raspberry Pi (for finetuning)

install docker (standard, not nvidia-docker)

start up slimmed docker container (smaller memory footprint, no jupyter, no gpu, 3.92GB)

docker build -t cs269_slim -f slim.dockerfile .
docker run -ti --shm-size=1g -e "TERM=xterm-256color" -v "$(pwd):/workspace" -p 7722:7722 -v /etc/localtime:/etc/localtime:ro cs269_slim

to ssh into container

ssh -o "UserKnownHostsFile=/dev/null" -o "StrictHostKeyChecking=no" root@localhost -p 7722

Instructions to run on Hoffman2

Hoffman2 does not support docker so you will need to manually set up the conda environment

Setting up the conda environment

qrsh -l h_rt=1:00:00,h_data=4G,gpu,RTX2080Ti

module load python/anaconda3 module load cuda/10.2 module load glibc/2.14

conda create --name EML conda activate EML

conda install pytorch torchvision -c pytorch -y conda config --append channels conda-forge

conda install expect gdown snakeviz -y pip install rigl_torch

Download jupyter notebook login script

Download login script

wget https://gitlab.idre.ucla.edu/dauria/jupyter-notebook/raw/master/h2jupynb

Logging in to conda environment from jupyter notebook

python h2jupynb -u username -v anaconda3 -d Edge_Meta_Learning -t session_length_in_hours -m session_memory_in_GB -g yes -c V100

Instructions to profile code

docker run ... python3 -m cProfile -o results.prof main.py
snakeviz results.prof -p 4433 -s -H 0.0.0.0
go to localhost:4433/snakeviz/%2Fworkspace%2Fresults.prof in browser