ultimate-utils 0.7.0

Brando's ultimate utils for science, machine learning and AI

Ultimate-utils

Ulitmate-utils (or uutils) is collection of useful code that Brando has collected through the years that has been useful accross his projects. Mainly for machine learning and programming languages tasks.

Installing Ultimate-utils

WARNING: YOU HAVE TO INSTALL PYTORCH ON YOUR OWN (WITH CUDA IF YOU NEED A GPU)

Standard pip install [Recommended]

If you are going to use a gpu the do this first before continuing (or check the offical website: https://pytorch.org/get-started/locally/):

pip install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu113 
pip install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu113 --upgrade

Otherwise, just doing the following should work.

pip install ultimate-utils

If that worked, then you should be able to import is as follows:

import uutils

note the import statement is shorter than the library name (ultimate-utils vs uutils).

Note, for an older version of uutils you might need to downgrade pytorch related stuff by doing:

pip uninstall torch torchvision torchaudio
pip install torch==1.9.1+cu111 torchvision==0.10.1+cu111 torchaudio==0.9.1 -f https://download.pytorch.org/whl/torch_stable.html

Manual installation [for Development]

To use uutils first get the code from this repo (e.g. fork it on github):

git clone [e-mail]:brando90/ultimate-utils.git

Then install it in development mode in your python env with python >=3.9 (read modules_in_python.md to learn about python envs). E.g. create your env with conda:

conda create -n uutils_env python=3.9
conda activate uutils_env

Then install uutils in edibable mode and all it's depedencies with pip in the currently activated conda environment:

pip install -e ~/ultimate-utils/ultimate-utils-proj-src

No error should show up from pip. To test the installation uutils do:

python -c "import uutils; uutils.hello()"
python -c "import uutils; uutils.torch_uu.hello()"

it should print something like the following:

hello from uutils __init__.py in:
<module 'uutils' from '/Users/brando/ultimate-utils/ultimate-utils-proj-src/uutils/__init__.py'>


hello from torch_uu __init__.py in:
<module 'uutils.torch_uu' from '/Users/brando/ultimate-utils/ultimate-utils-proj-src/uutils/torch_uu/__init__.py'>

To test (any) pytorch do:

python -c "import uutils; uutils.torch_uu.gpu_test_torch_any_device()"

output:

(meta_learning_a100) [miranda9@hal-dgx diversity-for-predictive-success-of-meta-learning]$ python -c "import uutils; uutils.torch_uu.gpu_test()"
device name: A100-SXM4-40GB
Success, no Cuda errors means it worked see:
out=tensor([[ 0.5877],
        [-3.0269]], device='cuda:0')
(meta_learning_a100) [miranda9@hal-dgx diversity-for-predictive-success-of-meta-learning]$ python -c "import uutils; uutils.torch_uu.gpu_test_torch_any_device()"
device name: A100-SXM4-40GB
Success, torch works with whatever device is shown in the output tensor:
out=tensor([[-1.9061],
        [ 1.3525]], device='cuda:0')

GPU TEST: To test if pytorch works with gpu do (it should fail if no gpus are available):

python -c "import uutils; uutils.torch_uu.gpu_test()"

output should be something like this:

(meta_learning_a100) [miranda9@hal-dgx diversity-for-predictive-success-of-meta-learning]$ python -c "import uutils; uutils.torch_uu.gpu_test()"
device name: A100-SXM4-40GB
Success, no Cuda errors means it worked see:
out=tensor([[ 0.5877],
        [-3.0269]], device='cuda:0')

[Adavanced] If using pygraphviz functions

If you plan to use the functions that depend on pygraphviz you will likely need to install graphviz first. On mac, brew install graphviz. On Ubuntu, sudo apt install graphviz.

Then install pygraphviz with

pip install pygraphviz

If the previous steps didn't work you can also try installing using conda (which seems to install both pygraphviz and graphviz`):

conda install -y -c conda-forge pygraphviz

to see details on that approach see the following stack overflow link question: https://stackoverflow.com/questions/67509980/how-does-one-install-pygraphviz-on-a-hpc-cluster-without-errors-even-when-graphv

To test if pygraphviz works do:

python -c "import pygraphviz"

Nothing should return if successful.

Contributing

Feel free to push code with pull request. Please include at least 1 self-contained test (that works) before pushing.

How modules are imported in a python project

Read the modules_in_python.md to have an idea of the above development/editable installation commands.

Executing tensorboard experiment logs from remote

• visualize the remote logs using pycharm and my code (TODO: have the download be automatic...perhaps not needed)

1. Download the code from the cluster using pycharm remote
2. Then copy paste the remote path (from pycharm, browse remote)
3. Using the copied path run tbb path2log e.g. tbbb /home/miranda9/data/logs/logs_Mar06_11-15-02_jobid_0_pid_3657/tb

to have tbbb work as the command add to your .zshrc (or .bashrc):

alias tbb="sh ${HOME}/ultimate-utils/run_tb.sh"

then the command tbb path2log should work.

ref: see files

• https://github.com/brando90/ultimate-utils/blob/master/run_tb.sh
• https://github.com/brando90/ultimate-utils/blob/master/ultimate-utils-proj-src/execute_tensorboard.py

Pushing to pypi

For full details see

~/ultimate-utils/tutorials_for_myself/pushing_to_pypi/README.md

For quick push do:

cd ~/ultimate-utils/
rm -rf build
rm -rf dist
cd ~/ultimate-utils/
python setup.py sdist bdist_wheel
twine check dist/*
twine upload dist/*

then:

cd ~/ultimate-utils/
rm -rf build
rm -rf dist

Testing pip install with docker

Create & pull image:

docker run -ti continuumio/miniconda3

then do

pip install ultimate-utils

it should be installed. You can import and test the print as mentioned above.

Citation

If you use this implementation consider citing us:

@software{brando2021ultimateutils,
    author={Brando Miranda},
    title={Ultimate Utils - the Ultimate Utils library for Machine Learning and Artificial Intelligence},
    url={https://github.com/brando90/ultimate-utils},
    year={2021}
}

A permanent link lives here: https://www.ideals.illinois.edu/handle/2142/112797

Docker container

If you have issues with docker & space run warning this will remove all your docker images:

docker system prune -af

To build the docker image form the computer that will run the docker container

# for adm/x86
docker pull brandojazz/ultimate-utils:test
# for adm/x86
docker pull brandojazz/ultimate-utils:test_arm

or do the standard docker build for the image for an x86 machine:

# for adm/x86
docker build -t brandojazz/ultimate-utils:test ~/ultimate-utils/
# for arm/m1
docker build -f ~/ultimate-utils/Dockerfile_arm -t brandojazz/ultimate-utils:test_arm ~/ultimate-utils/

docker login
# for adm/x86
docker push brandojazz/ultimate-utils:test
# for arm/m1
docker push brandojazz/ultimate-utils:test_arm

docker images

Run container:

docker run -ti brandojazz/ultimate-utils:test_arm bash

or in development mode:

docker run -v ~/ultimate-utils:/home/bot/ultimate-utils \
           -v ~/data:/home/bot/data \
           -ti brandojazz/ultimate-utils:test_arm bash

Inside the docker container:

python -c "import uutils; uutils.torch_uu.gpu_test_torch_any_device()"

Appendix

Data storage at UIUC vision cluster IC

home -> /home/miranda9 (export HOME=/home/miranda9, echo $HOME)
code/git -> home
ssh -> home
push on save -> home
.bashrc -> home
conda -> shared (miniconda)

data & logs & ckpts -> home (mkdir ~/data)
wandb -> home but delete folder after each run (rm -rf $WANDB_DIR)

My goal is to put the large heavy stuff (e.g. conda, data, ) at /shared/rsaas/miranda9/. Warning: due to the vpn if you run one of this commands and you lose connection you will have to do it again and might have half a transfer of files. So run them in a job.sub command or re pull them from git then do a soft link.

mv ~/data /shared/rsaas/miranda9
ln -s /shared/rsaas/miranda9/data ~/data 

mv ~/miniconda3 /shared/rsaas/miranda9
ln -s /shared/rsaas/miranda9/miniconda3 ~/miniconda3

mv ~/data_folder_fall2020_spring2021 /shared/rsaas/miranda9/
ln -s /shared/rsaas/miranda9/data_folder_fall2020_spring2021 ~/data_folder_fall2020_spring2021

# --

# mv ~/diversity-for-predictive-success-of-meta-learning /shared/rsaas/miranda9
cd /shared/rsaas/miranda9/
git clone [e-mail]:brando90/diversity-for-predictive-success-of-meta-learning.git 
# ln -s file1 link1
ln -s /shared/rsaas/miranda9/diversity-for-predictive-success-of-meta-learning ~/diversity-for-predictive-success-of-meta-learning 

mv ~/Does-MAML-Only-Work-via-Feature-Re-use-A-Data-Set-Centric-Perspective /shared/rsaas/miranda9
ln -s /shared/rsaas/miranda9/Does-MAML-Only-Work-via-Feature-Re-use-A-Data-Set-Centric-Perspective ~/Does-MAML-Only-Work-via-Feature-Re-use-A-Data-Set-Centric-Perspective 

mv ~/ultimate-anatome /shared/rsaas/miranda9
ln -s /shared/rsaas/miranda9/ultimate-anatome ~/ultimate-anatome 

mv ~/ultimate-aws-cv-task2vec /shared/rsaas/miranda9
ln -s /shared/rsaas/miranda9/ultimate-aws-cv-task2vec ~/ultimate-aws-cv-task2vec 

# mv ~/ultimate-utils /shared/rsaas/miranda9
cd /shared/rsaas/miranda9/
git clone [e-mail]:brando90/ultimate-utils.git
ln -s /shared/rsaas/miranda9/ultimate-utils ~/ultimate-utils 

mv ~/pycoq /shared/rsaas/miranda9
ln -s /shared/rsaas/miranda9/pycoq ~/pycoq 

mv ~/rfs /shared/rsaas/miranda9
ln -s /shared/rsaas/miranda9/rfs ~/rfs 

mv ~/automl-meta-learning /shared/rsaas/miranda9
ln -s /shared/rsaas/miranda9/automl-meta-learning ~/automl-meta-learning 

mv ~/wandb /shared/rsaas/miranda9
ln -s /shared/rsaas/miranda9/wandb ~/wandb 

cd /shared/rsaas/miranda9/
git clone [e-mail]:brando90/proverbot9001.git
ln -s /shared/rsaas/miranda9/proverbot9001 ~/proverbot9001 

to check real path (for soft links) do e.g. realpath /home.

SNAP servers

Logging in/sshing:

ssh [e-mail]
mkdir /dfs/scratch0/brando9
cd /dfs/scratch0/brando9
# ln -s file1 link1
#ln -s /dfs/scratch0/brando9 /afs/cs.stanford.edu/u/brando9/dfs/scratch0/brando9

# -- Titan X, 12 GB
ssh [e-mail]
ssh [e-mail]

# -- 2080, 11 GB
ssh [e-mail]
ssh [e-mail]
ssh [e-mail]
ssh [e-mail]

# -- A4000 16 GB [SK]
mercury1
mercury2

# -- RTX 8000 48GB, local storage 11TB
ssh [e-mail]
ssh [e-mail]
ssh [e-mail]
ssh [e-mail]

# -- a100 80GB, local storage 56TB
ssh [e-mail]
...
ssh [e-mail]

# -- a100 80GB, local storage ... [SK]
ssh brando9@...stanford.edu

My folder set up:

chmod a+x ~/diversity-for-predictive-success-of-meta-learning/main.sh
#chmod a+x /shared/rsaas/miranda9/diversity-for-predictive-success-of-meta-learning/main.sh

HOME -> /dfs/scratch0/brando9/ so that ckpting/logging works (wanbd logs to local lfs)
AFS -> /afs/cs.stanford.edu/u/brando9, export AFS=/afs/cs.stanford.edu/u/brando9, alias afs='cd $AFS' (where code is, need to fix data -> symlink to dfs data folder)
bashrc.user -> afs with symlynk to dfs (since HOME is at dfs), cat ~/.bashrc.user
ssh -> for now in afs because previous stuff broke when it was in dfs, need to test a little more
code/git -> code in /afs/cs.stanford.edu/u/brando9, so that push/sftp on save works in pycharm. Symlink them to afs so that they are visible at home=dfs/...
push on save -> root of projs /afs/cs.stanford.edu/u/brando9, make sure if you have root set automatically that you give the relative path on the deployment mapping (avoid putting root of hpc twice by accident)
wandb -> to local lfs of cluster, since that file really doesnt matter to me, just has to be somewhere so wandb works, see echo $LOCAL_MACHINE_PWD or/and ls $LOCAL_MACHINE_PWD, (if it fails put to dfs)

# conda ->  dfs and two back ups 
conda -> /dfs/scratch0/brando9 so any server has access to it, plus they are big so dont want to overwhelm afs (does symlinking conda to afs makes sense?), ls /dfs/scratch0/brando9/miniconda/envs & python -c "import uutils;uutils.get_home_pwd_local_machine_snap()" should work 

data -> /dfs/scratch0/brando9/ but with a symlink to /afs/cs.stanford.edu/u/brando9/data, TODO: https://intellij-support.jetbrains.com/hc/en-us/requests/4447850
# ln -s file1 link1
ln -s /afs/cs.stanford.edu/u/brando9/data /dfs/scratch0/brando9/data

Getting started:

# - create home
mkdir /dfs/scratch0/brando9
ls /dfs/scratch0/brando9

# note, export sets it for current shell and all subshells created
# - cd to home dfs
cd /dfs/scratch0/brando9
export HOME=/dfs/scratch0/brando9

# - to edit .bashrc.user
vim /afs/cs.stanford.edu/u/brando9/.bashrc.user
# ln -s file1 link1
ln -s /afs/cs.stanford.edu/u/brando9/.bashrc.user ~/.bashrc.user 

# - get hostname without stanford to set lfs for wandb
# export LOCAL_MACHINE_PWD=$(python -c "import uutils;uutils.get_home_pwd_local_machine_snap()")
export HOSTNAME=$(hostname)
export LOCAL_MACHINE_PWD="/lfs/${HOSTNAME::-13}/0/brando9"
echo LOCAL_MACHINE_PWD = $LOCAL_MACHINE_PWD

Installing conda:

# - in dfs when home is dfs
echo $HOME
wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh -O ~/miniconda.sh
bash ~/miniconda.sh -b -p $HOME/miniconda
ls -lah ~

export PATH="$HOME/miniconda/bin:$PATH"
echo $PATH
conda

source ~/miniconda/bin/activate
conda init
conda init bash
conda update -n base -c defaults conda
conda install conda-build

conda create -n metalearning_gpu python=3.9
conda activate metalearning_gpu
conda create -n iit_synthesis python=3.9
conda activate iit_synthesis
conda list

# - in afs when home is dfs but miniconda is installed to afs 
# WARNING (DISK QUOTA, DONT USE)
echo $AFS
wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh -O $AFS/miniconda.sh
bash $AFS/miniconda.sh -b -p $AFS/miniconda
ls -lah $AFS

export PATH="$AFS/miniconda/bin:$PATH"
echo $PATH
conda

source $AFS/miniconda/bin/activate
conda init
conda init bash
conda update -n base -c defaults conda
conda install conda-build

conda create -n metalearning_gpu python=3.9
conda activate metalearning_gpu
conda info -e

conda create -n iit_synthesis python=3.9
conda activate iit_synthesis
conda info -e

conda list

# - installing full anaconda
echo $HOME
wget https://repo.anaconda.com/archive/Anaconda3-2022.10-Linux-x86_64.sh -O ~/anaconda.sh
#wget https://repo.continuum.io/conda/Anaconda3-latest-Linux-x86_64.sh -O ~/anaconda.sh
#bash ~/anaconda.sh -b -p $HOME/anaconda
nohup bash ~/anaconda.sh -b -p $HOME/anaconda > anaconda_install.out &
tail -f anaconda_install.out
ls -lah ~

export PATH="$HOME/anaconda/bin:$PATH"
echo $PATH
conda

source ~/anaconda/bin/activate
conda init
conda init bash
conda update -n base -c defaults conda
conda install conda-build

conda create -n metalearning_gpu python=3.9
conda activate metalearning_gpu

nohup sh ~/diversity-for-predictive-success-of-meta-learning/install.sh > div_install.out &
tail -f div_install.out

nohup sh ~/diversity-for-predictive-success-of-meta-learning/install.sh > div_install_miniconda.out &
tail -f div_install_miniconda.out

conda create -n iit_synthesis python=3.9
conda activate iit_synthesis
conda list


# - making a backup
# cp -R <source_folder> <destination_folder>
nohup cp -R anaconda anaconda_backup &
nohup cp -R miniconda miniconda_backup &

# /dfs/scratch0/brando9
nohup cp -R /dfs/scratch0/brando9/anaconda /dfs/scratch0/brando9/anaconda_backup &
nohup cp -R /dfs/scratch0/brando9/miniconda /dfs/scratch0/brando9/miniconda_backup &


nohup pip install -e ~/ultimate-utils/ &
nohup pip install -e ~/diversity-for-predictive-success-of-meta-learning/ &

mkdir /lfs/madmax/0/brando9
nohup cp -R ~/anaconda /lfs/madmax/0/brando9/anaconda_backup &
nohup cp -R ~/miniconda /lfs/madmax/0/brando9/miniconda_backup &

# - in case you need the path if you broke it
export PATH=/afs/cs.stanford.edu/u/brando9/miniconda/bin:/afs/cs.stanford.edu/u/brando9/miniconda/bin:/afs/cs.stanford.edu/u/brando9/miniconda/condabin:/afs/cs.stanford.edu/u/brando9/miniconda/bin:/dfs/scratch0/brando9/miniconda/bin:/usr/local/cuda-11.1/bin:/home/miranda9/miniconda3/bin:/usr/kerberos/sbin:/usr/kerberos/bin:/afs/cs/software/sbin:/afs/cs/software/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin:/usr/games:/usr/local/games:/snap/bin:/dfs/scratch0/brando9/my_bins
# Use this command to check your afs quota and usage 
fs lq ~/
fs lq $AFS

# - try using pyenv
conda deactivate metalearning_gpu
virtualenv -p /usr/local/bin/python3.p metalearning_gpu_pyenv
source metalearning_gpu_pyenv/bin/activate
which pip

nohup sh ~/diversity-for-predictive-success-of-meta-learning/install.sh > div_install_pyenv.out &
tail -f div_install_pyenv.out

Git cloning your code & SHH keys:

echo $HOME
mkdir /afs/cs.stanford.edu/u/brando9/.ssh
cd /afs/cs.stanford.edu/u/brando9/.ssh
#touch ~/.ssh/id_ed25519
ssh-keygen -t ed25519 -C "[e-mail]"
# copy paste the bellow path (needs absolute path to $HOME unfortuantely)
### (no for now) /dfs/scratch0/brando9/.ssh/id_ed25519
# just enter send it to /afs/cs.stanford.edu/u/brando9/.ssh/id_ed25519
# press enter twice or type passphrase twice

eval "$(ssh-agent -s)"
### (no for now) ssh-add /dfs/scratch0/brando9/.ssh/id_ed25519
ssh-add /afs/cs.stanford.edu/u/brando9/.ssh/id_ed25519

# add public key to your github
### (no for now) cat /dfs/scratch0/brando9/.ssh/id_ed25519.pub
cat /afs/cs.stanford.edu/u/brando9/.ssh/id_ed25519.pub
# copy contents of terminal
# paste to your github keys under settings under SSH and GPG keys

gitclone projs:

git clone [e-mail]:brando90/ultimate-utils.git
git clone [e-mail]:brando90/diversity-for-predictive-success-of-meta-learning.git
git clone [e-mail]:brando90/pycoq.git
git clone [e-mail]:FormalML/iit-term-synthesis.git
git clone [e-mail]:brando90/proverbot9001.git
git checkout develop
mkdir /dfs/scratch0/brando9/data
mkdir /dfs/scratch2/brando9/data

/afs/cs.stanford.edu/u/brando9
/dfs/scratch0/brando9
# - DFS setup
# ln -s file1 link1
ln -s /afs/cs.stanford.edu/u/brando9/ultimate-utils /dfs/scratch0/brando9/ultimate-utils
ln -s /afs/cs.stanford.edu/u/brando9/diversity-for-predictive-success-of-meta-learning /dfs/scratch0/brando9/diversity-for-predictive-success-of-meta-learning 
ln -s /afs/cs.stanford.edu/u/brando9/pycoq /dfs/scratch0/brando9/pycoq  
ln -s /afs/cs.stanford.edu/u/brando9/iit-term-synthesis /dfs/scratch0/brando9/iit-term-synthesis 
ln -s /dfs/scratch0/brando9/data /afs/cs.stanford.edu/u/brando9/data 
ln -s /afs/cs.stanford.edu/u/brando9/proverbot9001 /dfs/scratch0/brando9/proverbot9001 

# lfs conda lfs -> dfs conda
# ln -s file1 link1
ln -s /dfs/scratch0/brando9/anaconda /dfs/scratch0/brando9/proverbot9001 

# LFS setup
# ln -s file1 link1
echo $HOME
ln -s /afs/cs.stanford.edu/u/brando9/ultimate-utils $HOME/ultimate-utils
ln -s /afs/cs.stanford.edu/u/brando9/diversity-for-predictive-success-of-meta-learning $HOME/diversity-for-predictive-success-of-meta-learning 
ln -s /afs/cs.stanford.edu/u/brando9/pycoq $HOME/pycoq  
ln -s /afs/cs.stanford.edu/u/brando9/iit-term-synthesis $HOME/iit-term-synthesis 
# ln -s /dfs/scratch0/brando9/data $HOME/data 
mkdir -p $HOME/data
ln -s /afs/cs.stanford.edu/u/brando9/proverbot9001 $HOME/proverbot9001 

Using kerberos tmux (https://ilwiki.stanford.edu/doku.php?id=hints:long-jobs): note:

• need to run krbtmux per server
• to reattach and other tmux commands use tmux prefix not krbtmux

# - run expt with krbtmux 
ssh ...

top -u brando9
tmux ls
pkill -9 tmux -u brando9; pkill -9 krbtmux -u brando9; pkill -9 reauth -u brando9; pkill -9 python -u brando9; pkill -9 wandb-service* -u brando9;

# - start krbtmux
krbtmux
reauth
cd diversity-for-predictive-success-of-meta-learning
#source ~/.bashrc.user
# now you have a krbtmux that won't die, so you can 0. perhaps running reauth inside the main_krbtmux.sh script 1. run a python job for each krbtmux or 2. run multiple python jobs inside this krbtmux session

# - run expt
sh main_krbtmux.sh &
# - end of run expt

tmux detach
tmux attach -t 0
# 

# --
# if the above doesn't work then I will have to have a script that sets up the environment variables correctly for me & then manually type the python main:
# sh main_setup_jobid_and_out_files.sh
# python main.py &
# sh echo_jobid_and_out_files.sh

# type password, later how to pass it from terminal automatically https://unix.stackexchange.com/questions/724880/how-to-run-a-job-in-the-background-using-tmux-without-interacting-with-tmux-lik

# note that settip up the jobid & outfile for err and stdout can be done within python too if the current sh main_krbtmux.sh & doesn't work and we move to python main.py &

# -- some useful tmux commands & keystrokes
tmux new -s mysession
tmux kill-session -t mysession
tmux switch -t <session name or number>
tmux attach -t <session number>
tmux ls
tmux detach

C-b ) = next session
C-b ( = previous session
C-b [ = to scroll history
C-b d = detach from tmux without closing/killing the session but return normally to the terminal :) 

todo: https://unix.stackexchange.com/questions/724880/how-to-run-a-job-in-the-background-using-tmux-without-interacting-with-tmux-lik

Using nohup for long running jobs in snap TODO: https://unix.stackexchange.com/questions/724902/how-does-one-send-new-commands-to-run-to-an-already-running-nohup-process-or-run

TODO bellow doesn't work
##nohup sh -c 'echo $SU_PASSWORD | /afs/cs/software/bin/reauth; python -u ~/diversity-for-predictive-success-of-meta-learning/div_src/diversity_src/experiment_mains/main_diversity_with_task2vec.py --manual_loads_name diversity_ala_task2vec_hdb1_mio > $OUT_FILE 2> $ERR_FILE' > $PWD/main.sh.nohup.out$SLURM_JOBID &

todo: perhaps opening a krbtmux and running commands there e.g. python main.sh & would work? (perhaps we don't even need nohup)

Using gpus snap: https://ilwiki.stanford.edu/doku.php?id=hints:gpu

source cuda11.1
# To see Cuda version in use
nvcc -V

Check disk space/quota:

df -h /dfs/scratch0

TODO, to reuse code, have something that checks the name of the cluster (if it's one of SNAPS, i.e. put the prefixes of snap ampere etc and run the code bellow, else do nothing let slurm, condor do it):

import os
os.environ["CUDA_DEVICE_ORDER"]="PCI_BUS_ID"
os.environ["CUDA_VISIBLE_DEVICES"]="0"

# likely as above, set HOME on the script you are running to log correctly

Creating folders in scratch with lots of space.

cd /dfs/scratch0/ 
mkdir brando9
cd /dfs/scratch0/brando9

check quota in afs

(metalearning_gpu) brando9~ $ fs lq $AFS
Volume Name                    Quota       Used %Used   Partition
user.brando9                 5000000    1963301   39%          6% 

weird bug activate doesn't work

export PATH="$HOME/miniconda/bin:$PATH"
echo $PATH
source ~/miniconda/bin/activate
conda activate metalearning_gpu
conda

#mv activate.c~ activate.c
#mv deactivate.c~ deactivate.c
#mv conda.c~ conda.c
#
#gcc -Wall activate.c -o activate
#gcc -Wall deactivate.c -o deactivate
#gcc -Wall conda.c -o conda

sh activate.c
sh deactivate.c
python conda.c

Seems current meta-learning uses:

torch                     1.9.1+cu111              pypi_0    pypi

let's try for iit

# pip3 install torch==1.9.1+cu111 torchvision==0.10.1+cu111 torchaudio==0.9.1 -f https://download.pytorch.org/whl/torch_stable.html
#pip install torch==1.9.1+cu111 torchvision==0.10.1+cu111 torchaudio==0.9.1 -f https://download.pytorch.org/whl/torch_stable.html
pip install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu113
pip install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu113 --upgrade

Installing ruby (snap)

mkdir ~/.rbenv
cd ~/.rbenv
git clone https://github.com/rbenv/rbenv.git .

export PATH="$HOME/.rbenv/bin:$PATH"
eval "$(rbenv init -)"
echo 'export PATH="$HOME/.rbenv/bin:$PATH"' >> ~/.bashrc.user
echo 'eval "$(rbenv init -)"' >> ~/.bashrc.user
#    exec $SHELL
bash

rbenv -v

# - install ruby-build
mkdir ~/.ruby-build
cd ~/.ruby-build
git clone https://github.com/rbenv/ruby-build.git .

export PATH="$HOME/.ruby-build/bin:$PATH"
echo 'export PATH="$HOME/.ruby-build/bin:$PATH"' >> ~/.bashrc.user
#    exec $SHELL
bash

ruby-build --version

# - install ruby without sudo -- now that ruby build was install
mkdir -p ~/.local
#    ruby-build 3.1.2 ~/.local/
rbenv install 3.1.2
rbenv global 3.1.2

ruby -v

Installation opam (snap)

Conda install failed so I did this one:

# - opam (snap, no sudo)
# ref: https://askubuntu.com/questions/339/how-can-i-install-a-package-without-root-access
apt-get download opam
#apt-get download opam_1.2.2-4_amd64
#ls | less
mkdir -p ~/.local
dpkg -x opam_1.2.2-4_amd64.deb ~/.local/bin
export PATH="$HOME/.local/bin:$PATH"
echo 'export PATH="$HOME/.local/bin:$PATH"' >> ~/.bashrc.user

tr ':' '\n' <<< "$PATH"

opam --version

LFS

idea2: use lfs, simplest
- repos: symlink lfs -> afs 
- conda symlink: lfs -> afs
    - later, migrate to miniconda (over winter break)
- data: lfs -> dfs (effortless)
- my bins/locals: in lfs
    - opam
    - ruby
- deployment path: to ampere4 -> dfs (effortless)
    - since data is pointing to dfs (effortless) nothing to be done
    - once it changes to lfs then we change the deployment path
    - only need hyperion1 to work for repos/code sending since code is in afs & lfs -> afs for those on the snap config
- [sounds confusing, solving a problem I don't have, maybe creating the data sets to dfs is fine
	- only need data in pycharm to point to lfs (not dfs)]

# ln -s file1 link1
ln -s /afs/cs.stanford.edu/u/brando9/.bashrc.user /dfs/scratch0/brando9/.bashrc.user
ln -s /afs/cs.stanford.edu/u/brando9/.bashrc.lfs /dfs/scratch0/brando9/.bashrc.lfs

CPU

ref: https://ilwiki.stanford.edu/doku.php?id=snap-servers:snap-servers#compute_servers