Repolode
Submit
togethercomputer

togethercomputer/together-cookbook

A collection of Jupyter notebooks and guides showing how to build with open-source models on the Together AI platform.

View on GitHub
Stars
1,194
Forks
221
Open issues
7
Last commit
4 days ago

What is it?

Together Cookbook is a set of runnable recipes for developers who want to build with open-source models hosted on Together AI. Most entries are Jupyter notebooks that call the Together API, and the code is meant to be copied into your own projects. A Together AI API key is needed to run the examples, and the Python and JavaScript samples can be adapted to other languages that can call the API.

The recipes cover chat and text inference, vision tasks such as OCR, grounding, document parsing and video understanding, image generation, and structured outputs. They also include agent workflow patterns and integrations with frameworks like LangGraph, DSPy, PydanticAI and Agno, plus fine-tuning jobs (DPO, continual, long-context, vision-language), evaluation workflows such as LLM-as-a-judge, and a Grafana dashboard for monitoring dedicated endpoints.

The folder layout mirrors Together AI's product areas, and notebooks that rely on retired models are moved to an archived folder.

Key features

  • 01Notebooks cover chat, vision, image generation, structured outputs and batch inference.
  • 02Agent workflow examples integrate LangGraph, DSPy, PydanticAI, Agno and several tool providers.
  • 03Fine-tuning recipes include DPO, continual, long-context, function-calling and vision-language training.
  • 04Evaluation examples cover classification, model comparison, prompt testing and LLM judge tuning.
  • 05Each notebook has a Colab link so it can run without local setup.

Use cases

  • Prototyping a document OCR or multi-page PDF Q&A pipeline on open-source vision models.
  • Building a retrieval or multi-step agent on Together models with LangGraph or PydanticAI.
  • Fine-tuning an open model on preference data or function-calling examples.
  • Setting up LLM-as-a-judge evaluations to compare prompts or models.

Quick start

  1. 1

    Sign up for a Together AI API key

  2. 2

    Open a cookbook notebook in Colab or clone the repo and run its Jupyter notebook

Things to know

  • Examples require a Together AI API key and call Together's hosted models.
  • It is a collection of notebooks and guides, not an installable library.

In collections

Open-source AI agent frameworks & apps

Topics

#llm#open-source-models#jupyter-notebook#fine-tuning#agents#multimodal#evaluation#cookbook

FAQ

What is together-cookbook?
A collection of Jupyter notebooks and guides showing how to build with open-source models on the Together AI platform. Together Cookbook is a set of runnable recipes for developers who want to build with open-source models hosted on Together AI. Most entries are Jupyter notebooks that call the Together API, and the code is meant to be copied into your own projects. A Together AI API key is needed to run the examples, and the Python and JavaScript samples can be adapted to other languages that can call the API.
Is together-cookbook free to use? What license does it have?
together-cookbook is open source under the MIT license. Check the license terms before using it in commercial products.
Is together-cookbook still maintained?
Actively maintained. The last commit to together-cookbook was on Nov 3, 2026.
What can I use together-cookbook for?
Prototyping a document OCR or multi-page PDF Q&A pipeline on open-source vision models. Building a retrieval or multi-step agent on Together models with LangGraph or PydanticAI. Fine-tuning an open model on preference data or function-calling examples. Setting up LLM-as-a-judge evaluations to compare prompts or models.
How do I install together-cookbook?
Sign up for a Together AI API key Open a cookbook notebook in Colab or clone the repo and run its Jupyter notebook
Where can I get together-cookbook?
The source code and installation instructions are on GitHub: https://github.com/togethercomputer/together-cookbook

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects