:arrow_down: Download AI Workbench • :book: Read the Docs • :open_file_folder: Explore Example Projects • :rotating_light: Facing Issues? Let Us Know!
# Project Description This is an [NVIDIA AI Workbench](https://www.nvidia.com/en-us/deep-learning-ai/solutions/data-science/workbench/) project for bringing your own instance to Kaggle competitions. Users can use the Kaggle API to connect, download datasets from, and submit results to the Kaggle platform, providing a new way to seamlessly work locally on hardware of the user's choice, free from the limitations of cloud-based platforms like Google Colab and the Kaggle Kernel. | :warning: Heads Up | | :----------------------------- | | This project is compatible with Kaggle competitions that are reachable for submission via the Kaggle API. This project will **not** submit properly for competition formats that restrict users to the platform for a balanced hardware allowance requirement (like the "code" competition format). More details [here](https://www.kaggle.com/docs/competitions#competition-formats). | This project builds out a mirrored Kaggle environment and consists of the following notebooks to provide a seamless integration into the Kaggle competition platform: * ``01-data.ipynb``: This notebook walks the user through downloading competition datasets directly from the Kaggle platform. Simply type the competition name and get started! * ``02-code.ipynb``: This notebook walks through sample code for solving the end-to-end example competition started in ``01-data.ipynb`` and compiles a submission file. * ``03-submit.ipynb``: This notebook submits your outputs to the Kaggle competition. See how you stack up on the leaderboard! In addition to providing a seamless local experience for working with Kaggle competitions, AI Workbench also provides the following to users: * Easy version control and tracking of code via Github/Gitlab - say goodbye to manual versioning or CI/CD pipelines. * Get the advantages of using a local, dedicated IDE: robust debugging, intelligent code completion, and downloadable extensions. * Automatically set up and customize your local environment in minutes. * Plug into and access existing data sources locally without needing to upload them to third parties. * No Internet? No problem. Develop while offline! | :memo: Remember | | :---------------------------| | This project is meant as an example workflow and a starting point; you are free to swap out the example competition, add new datasets and models, rearrange the interface, or edit the source code as you see fit! | ## Sizing Guide | GPU VRAM | Example Hardware | Compatible? | | -------- | ------- | ------- | | N/A | CPU-only | Y | | <16 GB | RTX 3080, RTX 3500 Ada | Y | | 16 GB | RTX 4080 16GB, RTX A4000 | Y | | 24 GB | RTX 3090/4090, RTX A5000/5500, A10/30 | Y | | 32 GB | RTX 5000 Ada | Y | | 40 GB | A100-40GB | Y | | 48 GB | RTX 6000 Ada, L40/L40S, A40 | Y | | 80+ GB | A100-80GB | Y | # Quickstart ## Prerequisites AI Workbench will prompt you to provide a few pieces of information before running any apps in this project. Ensure you have this information ready. * A Kaggle Username. You can find this in ``kaggle.json`` when you click "Create New Token" [here](https://www.kaggle.com/settings). * A Kaggle API Key. You can find this in ``kaggle.json`` when you click "Create New Token" [here](https://www.kaggle.com/settings). Alternatively, if you prefer working directly with the ``kaggle.json`` credential file, you can delete the above secrets and add your ``kaggle.json`` file directly at the top level of this project repository. However, you may need to run a ``chmod 600 /path/to/kaggle.json`` to prevent other users from reading your file. ## Tutorial (Desktop App) If you do not NVIDIA AI Workbench installed, first complete the installation for AI Workbench [here](https://www.nvidia.com/en-us/deep-learning-ai/solutions/data-science/workbench/). | :bulb: Tip | | :-----------------------| | Working in the AI Workbench command-line interface (CLI)? Skip to the [next section](#tutorial-cli) for a CLI-only tutorial! | Let's get started! 1. Fork this Project to your own GitHub namespace and copy the link ``` https://github.com/[your_namespace]/