Installing and setting up the CLI
Iskra works with any Python 3 project, whether it's a Conda, venv or global enviornment. We strongly recommend creating local environments for your projects (you can read about Conda here and venv here).
In order to install the CLI, activate your environment and type
pip install iskra-cli
If the installation was sucessful, running
iskra should return a list of available instructions. If you execute any of these commands, you will be guided through authentication account creation and billing configuration accordingly.
If you are not authenticated, the CLI will ask you for your e-mail address and we will send you an activation code. The file containing your token will be located in
~ is your home folder). This file consists access information to Iskra platform, so don't share it.
If you haven't used Iskra in the current working directory, it will create
.iskra.json file that stores basic information about your project, such as project ID. You can safely commit this file to your repository.
If you have not set up your billing account, you will be asked to provide your billing details. Don't worry - we won't charge you anything yet, as until the end of March, Iskra is free for all use. Please see the pricing section.
If you attempt an action that requires authenticated user, Iskra will ask you to login anyway, but you can log in by yourself aswell with
Expectedly, in order to log out, type
Starting a training job
iskra train your_training_script.py
your_training_script.py can be any valid Python 3 script. Our system automatically resolves all the dependencies based on the local environment you are using, unless you specify the
--requirements flag which points to a
requirements.txt file (more on how to build one here).
Welcome to Iskra, can you tell us your email: email@example.com Click in the link we sent you via email... Ok, all setup...yolov3-tf2 (bd878b07-bb1a-447f-a4ba-ce87de28df9d) https://app.iskra.ml/project/bd878b07-bb1a-447f-a4ba-ce87de28df9d :: Introspecting environment :: Waiting for remote machine to be ready :: Synchronizing source code :: Building remote environment <OUTPUT OF YOUR TRAINING HERE>
After starting the training, the terminal will output the process of creation of remote environment and logs output from your training script. Pressing CTRL+C or interrupting the process in any other way will not stop the training.
Stopping a training
iskra stop <training_job_id>
Retrieving the recent 5 training jobs
Executing this command in a working directory that is an Iskra project will return the last 5 trainings including stopped/failed jobs, along with their IDs.
Project: yolov3-tf2 (bd878b07-bb1a-447f-a4ba-ce87de28df9d) Running Time CPU GPU qg8MswBFfYO5pTW49lIX 12 min 65% 99% pvMnoJbg9iUTP576XOLt 45 seg 65% 99% Stopped Time Airtefacts size FgcASyKTBvloEWde7UnH 12 hours 632 Mb
Logs of a training job
iskra logs <training_job_id>
If you would like to display logs in real time, you can add
iskra logs -f <training_job_id>
Downloading training job artifacts
This will open the browser in the curren tproject and there you can navigate to an specific training job and be able to browse and download its files.
Creating a dataset
In a given folder, execute
iskra dataset create --name
This will create a dataset in Iskra, create a configuration file in your filesystem, and upload the folders contents to our servers.
Synchronizing a dataset
If you would like to later change the contents of your dataset and upload these changes to Iskra, simply run
iskra dataset sync
Deleting a dataset
If you would like to delete a dataset from Iskra, run
iskra dataset delete
This, except from dataset configuration file, will not any remove files in your local filesystem.