Getting Started


There are two things you will need to do before you start working with

After this, you're free to explore the platform and it's functionality. As a good starting point, we've included a section about development on GPU with Jupyter Notebooks.

Installing the CLI

Web Terminal doesn't require installation and can quickly get you familiar with, allowing you to work with the platform in a browser.

However, installing CLI locally may prove more effective for long-term use:

  • You won't need to pay for simply running the job like you do in Web UI.

  • Your source code and other local files will be saved directly on your machine.

Installation instructions

Installing via pipx

Our neuro-all package available in pipx will automatically install all required components:

$ pip install pipx
$ pipx install neuro-all
$ pipx upgrade neuro-all

Installing via pip

You can also install all of the components through pip. CLI requires Python 3 installed (recommended: 3.8; required: 3.7.9 or newer). We suggest using the Anaconda Python 3.8 Distribution.

$ pip install -U neuro-cli neuro-extras neuro-flow
$ neuro login

If your machine doesn't have GUI, use the following command instead of neuro login:

$ neuro config login-headless

Understanding the core concepts

On the Core level, you will work with jobs, environments, and storage. To be more specific, a job (an execution unit) runs in a given environment (Docker container) on a given preset (a combination of CPU, GPU, and memory resources allocated for this job) with several storage instances (block or object storage) attached.

Here are some examples.

Hello, World!

Run a job on CPU which prints “Hello, World!” and shuts down:

$ neuro run --preset cpu-small --name test ubuntu -- echo Hello, World!

Executing this command will result in an output like this:

√ Job ID: job-7dd12c3c-ae8d-4492-bdb9-99509fda4f8c
√ Name: test
- Status: pending Creating
- Status: pending Scheduling
- Status: pending ContainerCreating
√ Http URL:
√ The job will die in a day. See --life-span option documentation for details.
√ Status: succeeded
√ =========== Job is running in terminal mode ===========
√ (If you don't see a command prompt, try pressing enter)
√ (Use Ctrl-P Ctrl-Q key sequence to detach from the job)
Hello, World!

A simple GPU job

Run a job on GPU in the default environment (neuromation/base) that checks if CUDA is available in this environment:

$ neuro run --preset gpu-small --name test neuromation/base -- python -c "import torch; print(torch.cuda.is_available());"

We used the gpu-small preset for this job. To see the full list of presets you can use, run the following command:

$ neuro config show

Working with platform storage

Create a new demo directory in the root directory of your platform storage:

$ neuro mkdir -p storage:demo

Run a job that mounts the demo directory from platform storage to the /demo directory in the job container and creates a file in it:

$ neuro run --volume storage:demo:/demo:rw ubuntu -- bash -c "echo Hello >> /demo/hello.txt"

Check that the file you have just created is actually on the storage:

$ neuro ls storage:demo

Developing on GPU with Jupyter Notebooks

Development in Jupyter Notebooks is a good example of how the Neuro Platform can be used. While you can run a Jupyter Notebooks session in one command through CLI or in one click in the web UI, we recommend project-based development. To simplify the process, we provide a project template which is based on the cookiecutter package and is a part of the Toolbox. This template provides the basic necessary folder structure and integrations with several recommended tools.

Initializing a Neuro cookiecutter project

First, you will need to install the cookiecutter package via pip or pipx:

$ pipx install cookiecutter

Now, to initialize a new Neuro cookiecutter project, run:

$ cookiecutter gh:neuro-inc/cookiecutter-neuro-project --checkout release

This command will prompt you to enter some info about your new project:

project_name [Neuro Project]: New Cookiecutter Project
project_dir [new cookiecutter project]:
project_id [new_cookiecutter_project]:
code_directory [modules]:
preserve Neuro Flow template hints [yes]:

Default values are indicated by square brackets [ ]. You can use them by pressing Enter.

To navigate to the project directory, run:

$ cd new-cookiecutter-project

Project structure

The structure of the project's folder will look like this:

├── .github/            <- Github workflows and a dependabot.yml file
├── .neuro/             <- neuro and neuro-flow CLI configuration files
├── config/             <- configuration files for various integrations
├── data/               <- training and testing datasets (we don't keep it under source control)
├── notebooks/          <- Jupyter notebooks
├── modules/            <- models' source code
├── results/            <- training artifacts
├── .gitignore          <- default .gitignore file for a Python ML project
├── .neuro.toml         <- autogenerated config file for Neuro CLI
├── .neuroignore        <- a file telling Neuro CLI which files to ignore while uploading to the platform storage
├──             <- autogenerated template reference
├──           <- autogenerated informational file
├── Dockerfile          <- description of the docker image used for training in your project
├── apt.txt             <- list of system packages to be installed in the training environment
├── requirements.txt    <- list of Python dependencies to be installed in the training environment
├── setup.cfg           <- linter settings (Python code quality checking)
└──   <- script used to update neuro-flow actions in one of the GitHub workflows

The template contains the neuro/live.yaml configuration file for neuro-flow. This file guarantees a proper connection between the project structure, the base environment that we provide, and actions with storage and jobs. For example, the upload command synchronizes sub-folders on your local machine with sub-folders on the persistent platform storage, and those sub-folders are synchronized with the corresponding sub-folders in job containers.

Setting up the environment and running Jupyter

To set up the project environment, run:

$ neuro-flow build train
$ neuro-flow mkvolumes

When these commands are executed, system packages from apt.txt and pip dependencies from requirements.txt are installed to the base environment. It supports CUDA by default and contains the most popular ML/AI frameworks such as Tensorflow and Pytorch.

For Jupyter Notebooks to run properly, the script and the notebook itself should be available on the storage. Upload the code directory containing this file to the storage by using the following command:

$ neuro-flow upload ALL

Now you need to choose a preset on which you want to run your Jupyter jobs. To view the list of presets available on the current cluster, run:

$ neuro config show 

To start a Jupyter Notebooks session run:

$ neuro-flow run jupyter

This command will open Jupyter Notebooks interface in your default browser.

Also, you can adjust the jupyter job's preset configuration by specifying the preset argument to reflect your preferred preset:

    action: gh:neuro-actions/jupyter@<version>
      preset: <gpu-preset-name>

After this, each time you run a Jupyter job, it will use the specified by default without the need for you to provide it in a CLI command:

$ neuro-flow run jupyter 

You can find more information about job description arguments here

Now, when you edit notebooks, they are updated on your platform storage. To download them locally (for example, to save them under a version control system), run:

$ neuro-flow download notebooks

Don’t forget to terminate your job when you no longer need it (the files won’t disappear after that):

$ neuro-flow kill jupyter

To check how many credits you have left, run:

$ neuro config show

Last updated