• Use VSCode or any IDE that supports Jupyter notebooks.
• Select the newly installed med-mnist environment as the kernel.
• Run the notebook cells.

Public/Private SSH Key Setup

• Generate an SSH key on your local machine (if you don’t have one).
• Add the public key to your RunPod/Vast.ai account settings.

Create a New Instance

• Pick a normal Linux image (for example, ubuntu:noble-20250127).
• Specify the following on-start script in the instance settings:

#!/usr/bin/env bash

# Usage:
# 1. Copy this script to the On-start Script section
# 2. Start the VM and wait for the setup to complete

set -eux

apt-get update

##
# 1. Install Miniconda
##
mkdir -p ~/miniconda3
wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh -O ~/miniconda3/miniconda.sh
bash ~/miniconda3/miniconda.sh -b -u -p ~/miniconda3
rm ~/miniconda3/miniconda.sh

# Activate Miniconda for this session
source ~/miniconda3/bin/activate

# Initialize Conda for all shell types
~/miniconda3/bin/conda init --all

# Ensure Conda is available in future shell sessions
# echo 'export PATH="$HOME/miniconda3/bin:$PATH"' >> ~/.bashrc
# echo 'source ~/miniconda3/bin/activate' >> ~/.bashrc

##
# Install Unison for File Sync
##
apt-get update
DEBIAN_FRONTEND=noninteractive apt-get install -y unison

##
# Disable auto tmux
##
touch ~/.no_auto_tmux

##
# Verify Installations
##
echo "Miniconda Installed:"
conda --version

echo "Unison Installed:"
unison -version

echo "SSH Status:"
systemctl status ssh | grep Active

echo "Setup Complete! Miniconda and Unison are ready."

Syncing Your Code with Unison
From your local machine, use a Unison profile or a single command:

Profile Example (unison_medmnist.prf):

 # Define local and remote sync locations
 root = <PATH>\med-mnist
 root = ssh://<USER>@<IP>:<PORT>//root/med-mnist

 # Sync the entire folder except ignored files
 ignore = Path datasets
 ignore = Path src/__pycache__
 ignore = .git

 # Ignore specific files
 ignore = Name README.md
 ignore = Name unison.prf
 ignore = Name .gitignore
 ignore = Name *.log

Copy unison_medmnist.prf to your local Unison directory (for example, C:\Users\<User>\.unison\) and run:

unison unison_medmnist.prf -auto -batch -repeat 2

One-Liner Command (avoids using a .prf file):

 unison "<PATH>\med-mnist" "ssh://<USER>@<IP>:<PORT>//root/med-mnist" -auto -batch -repeat 2 -ignore "Path datasets" -ignore "Path src/__pycache__" -ignore "Path .git" -ignore "Name README.md" -ignore "Name unison.prf" -ignore "Name .gitignore" -ignore "Name *.log"

SSH and Port Forwarding

ssh -p <PORT> <USER>@<IP> -L 8080:localhost:8080

The -L 8080:localhost:8080 flag forwards port 8080 on your local machine to port 8080 on the remote machine. This lets you connect to services (like Jupyter) running remotely on port 8080 by visiting localhost:8080 in your local browser.

Create and Activate Conda Environment

cd /root/med-mnist
conda env create -f environment.yml
conda activate med-mnist

Install Jupyter (if missing)

conda install jupyter

Run Jupyter Server on the Remote

jupyter notebook --no-browser --port 8080 --allow-root

Copy the localhost URL with the token from the terminal output.
Example: http://127.0.0.1:8080/?token=<SOME_TOKEN>

Checking Running Jupyter Notebooks and Tokens If you need to retrieve the Jupyter notebook token, then from another SSH session, run:

jupyter notebook list

This will show active Jupyter sessions along with their URLs and tokens.

Connect from VS Code or Another IDE

• Open your notebook on your local machine.
• Select “Remote Jupyter Server” and provide http://127.0.0.1:8080/?token=<SOME_TOKEN> as the server URL.
• Run the notebook cells as if it were local.