Fondant 0.8: Simplification, Sagemaker, RAG, and more!#
Hi all, we released Fondant 0.8, which brings some major new features and improvements:
- π We simplified and improved the way datasets are stored and accessed
- π The interface to compose a Fondant pipeline is now simpler and more powerful
- π AWS SageMaker is now supported as an execution framework for Fondant pipelines
- π The Fondant explorer was improved, especially for text and document data
- π We released a RAG tuning repository powered by Fondant
Read on for more details!
π We simplified and improved the way datasets are stored and accessed#
We listened to all your feedback and drastically simplified Fondant datasets, while solving some longstanding issues as part of the design.
Most important for you is that we flattened the datasets, removing the concept of subsets
from
Fondant. Which means you can now access the data fields directly!
π The interface to compose a Fondant pipeline is now simpler and more powerful.#
You can now chain components together using the read()
, apply()
and write
methods, removing
the need for specifying dependencies separately, making composing pipelines a breeze.
Some of the benefits of this new interface are:
- Support for overriding the produces and consumes of a component, allowing you to easily change the output of a component without having to create a custom
fondant_component.yaml
file. - We unlock the future ability to enable eager execution of components and interactive development of pipelines. Keep an eye on our next releases!
If you want to know more or get started you can check out the documentation
π AWS SageMaker is now supported as an execution framework for Fondant pipelines.#
You can now easily run your Fondant pipelines on AWS SageMaker using the fondant run sagemaker <pipeline.py>
command. Run fondant run sagemaker --help
to see the possible configuration options or check out the documentation.
πFondant explorer improvements#
We added a lot of improvements to the Fondant explorer, including:
- A pipeline overview showing the data flow through the pipeline
- A document viewer to inspect data (handy for RAG use cases)
- Better filtering, sorting and searching of data while exploring
To get started with the Fondant explorer, check out the documentation.
π We released a RAG tuning repository powered by Fondant#
This repository helps you tune your RAG system faster and achieve better performance using Fondant. Find the repository including a full explanation here.
It includes:
- A Fondant pipeline to ingest the data
- A Fondant pipeline to evaluate the data
- Multiple notebooks to go from a basic RAG pipeline to fully auto-tuned RAG pipelines
π§ New reusable RAG components#
A lot of new reusable components were added to the Fondant registry, letting you build new RAG pipelines quickly!
- Weaviate indexing components
- Qdrant indexing
- Ragas evaluation
- LlamaHub loading
- LangChain chunking and embedding
You can see some of these components in action in the RAG tuning repository.
π Install it now!#
And let us know what you think!