A lot of ML/AI workloads require powerful compute in order to process data at scale, run large language models, or fine-tune models on your own data. It is also important that these workloads are reproducible and easily shared with your team. Oxen.ai allows you to spin up Marimo Notebooks on a GPU with just a few clicks, without having to manage any infrastructure.

Select the GPU you want to use, and within a few seconds you will have a Notebook open with a GPU ready to go.

​

Example Notebooks

Notebooks give you the flexibility to build any workflow that you need, whether it is viewing data, labeling data, training models or running inference on large datasets. Below Coming soon you can find some examples to spark your imagination and get going.

• 🗺️ Explore, Process, and Version Data
• 🏷️ Build a Custom Labeling Tool
• 🔎 Compute Text Embeddings
• 🧪 Generate Synthetic Datasets
• 💬 Build a Chatbot Data Flywheel
• 📑 Process PDFs
• 🏋️‍♀️ Train a LLM
• 🕵️‍♀️ Evaluate an LLM

If you want to contribute more examples, feel free to make a pull request to [https://github.com/Oxen-AI/docs]

​

What is Marimo?

Marimo Notebooks are a reactive Python development environment. The Notebooks consist of cells of arbitrary Python code and can even contain UI elements.

Notebooks are great for exploring and cleaning data, evaluating and training models, or even building internal apps and interfaces for less technical people on the team to view or label data.

Notebooks can be run in edit mode, where you can see the code, and app mode where you only see the UI elements. Marimo comes out of the box with a lot of great UI elements, but you can also build your own.

The Notebooks are simply stored as Python code, without any fancy json configuration under the hood. Meaning the diffs are easy to grok, and anything you can write in raw Python can be turned into a Notebook.

​

Versioning Code, Data, and Models

As you are writing code in your Notebooks, you can commit the Notebook directly to your Oxen.ai repository along side your data and models. When you are ready, you can commit the Notebook to your repository to share with your team.

Since Oxen.ai is built to handle many files and large files, there is no reason to separate your code from your datasets and model weights. Store all three in your Oxen.ai repository, and share with your team to have reproducible pipelines.