Introduction

There are two things you will need to do before you start working with Neu.ro:

1. ​Install the CLI client.

2. ​Understand the platform's core concepts.

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 Neu.ro, allowing you to work with the platform in a browser.

However, installing Neu.ro CLI locally may prove more effective for long-term use.

Installation instructions

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

$ neuro config login-headless

$ conda install -c conda-forge make
$ conda install -c conda-forge git    
$ pip install -U neuro-cli neuro-extras neuro-flow
$ pip install -U certifi
$ neuro login

Understanding the core concepts

On the Neu.ro 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
√ Http URL: https://test--jane-doe.jobs.neuro-compute.org.neu.ro
√ 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 Neu.ro environment (neuromation/base) that checks if CUDA is available in this environment:

$ neuro run --preset gpu-k80-small --name test neuromation/base python -c "import torch; print(torch.cuda.is_available());"

We used the gpu-k80-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 --preset cpu-small --name test --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 a part of the Neu.ro Toolbox. This template provides the basic necessary folder structure and integrations with several recommended tools.

Initializing a project

To initialize a new project from the template, run:

$ neuro project init

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

project_name [Name of the project]: Neuro Tutorial
project_slug [neuro-tutorial]:
code_directory [modules]:

To navigate to the project directory, run:

$ cd neuro-tutorial

Project structure

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

neuro-tutorial
├── .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 for a Python ML project
├── .neuro.toml         <- autogenerated config file
├── HELP.md             <- autogenerated template reference
├── README.md           <- autogenerated informational file
├── Dockerfile          <- description of the base image used for 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)

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 myimage
$ 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 train.py script should be available on the storage. Upload the code directory containing this file to the storage by using the following command:

$ neuro-flow upload code

To start a Jupyter Notebooks session on GPU, run:

$ neuro-flow run jupyter

This command open Jupyter Notebooks interface in your default browser.

Now, when you edit notebooks, they are updated on your platform storage. To download them locally (for example, to connect them to 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-quota