Skip to content

Releases: ml6team/fondant

0.8.dev6

12 Dec 18:57
bc7bc98
Compare
Choose a tag to compare
0.8.dev6 Pre-release
Pre-release

What's Changed

Full Changelog: 0.8.dev5...0.8.dev6

0.8.dev5

12 Dec 08:32
589e327
Compare
Choose a tag to compare
0.8.dev5 Pre-release
Pre-release

What's Changed

Full Changelog: 0.8.dev4...0.8.dev5

0.8.dev4

07 Dec 08:55
dac4e20
Compare
Choose a tag to compare
0.8.dev4 Pre-release
Pre-release

What's Changed

Full Changelog: 0.8.dev3...0.8.dev4

0.8.dev3

04 Dec 13:47
2b47ba0
Compare
Choose a tag to compare
0.8.dev3 Pre-release
Pre-release

What's Changed

New Contributors

  • @NSFF made their first contribution in #694

Full Changelog: 0.8.dev2...0.8.dev3

0.8.dev2

30 Nov 15:24
7ba372f
Compare
Choose a tag to compare
0.8.dev2 Pre-release
Pre-release

What's Changed

Full Changelog: 0.8.dev1...0.8.dev2

0.8.dev1

27 Nov 13:20
8b54505
Compare
Choose a tag to compare
0.8.dev1 Pre-release
Pre-release

What's Changed

Full Changelog: 0.8.dev0...0.8.dev1

0.8.dev0

27 Nov 12:12
f87217a
Compare
Choose a tag to compare
0.8.dev0 Pre-release
Pre-release

What's Changed

New Contributors

Full Changelog: 0.7.0...0.8.dev0

0.7.0

20 Nov 14:45
81259cc
Compare
Choose a tag to compare

Highlights

  • We restructured and updated our documentation, which should make it easier to get started, and learn more advanced concepts as you go.

    👉 Check it out at fondant.ai!

  • We moved our example pipelines into separate repositories, which will make it easier to get started with them:

    • 📖 RAG ingestion pipeline
      An end-to-end Fondant pipeline that prepares documents for a RAG (Retrieval Augmented Generation) system by chunking and embedding them, and writing them to a vector store.

    • 🛋️ ControlNet Interior Design Pipeline
      An end-to-end Fondant pipeline to collect and process data for the fine-tuning of a ControlNet model, focusing on images related to interior design.

    • 🖼️ Filter creative common license images
      An end-to-end Fondant pipeline that starts from our Fondant-CC-25M creative commons image dataset and filters and downloads the desired images.

    • 🔢 Datacomp pipeline
      An end-to-end Fondant pipeline filtering image-text data to train a CLIP model for the DataComp competition.

  • We split our component and pipeline SDK, so only the actual required dependencies are installed.

    In components, install the component extra so you can use the fondant.component SDK.

    pip install fondant[component]

    Locally, just install Fondant without extras to fondant.pipeline SDK and CLI.

    For the local runner:

    pip install fondant

    Or with the appropriate extra for your specific runner:

    pip install fondant[vertex]

    More info on the available installation options is available here.

All changes

New Contributors

Full Changelog: 0.6.2...0.7.0

0.6.2

20 Oct 09:28
f6d8753
Compare
Choose a tag to compare

Make docker connection during fondant build command more robust on Mac.

What's Changed

Full Changelog: 0.6.1...0.6.2

0.6.1

19 Oct 13:47
f8f4946
Compare
Choose a tag to compare

This is a re-packaged release of version 0.6.1.

Version 0.6.0 packaged an older commit due to a bug in our release system.

Highlights

  • Vertex AI is now supported as a backend for pipeline execution.

    Simply run fondant run vertex <pipeline.py> to submit your pipeline.
    Run fondant run vertex --help to see the possible configuration options.

  • The reusable components are now available on DockerHub under the fndnt organization.

    DockerHub is supported more broadly than Github container registry which we were using before.

  • Previously executed components are now cached when re-executed with the same arguments.

    • This makes it easier to iterate on development of down-stream components
    • This allows you to resume failed pipelines from their failed step
  • Added fondant build command which let's you build fondant components easily

    Run fondant build <component_dir>. Check fondant build -h for options.
    The command will also update the image reference in the fondant_component.yaml to the newly built one.

  • We migrated from KfP v1 to KfP v2. This means:

    • We now benefit from the latest KfP developments
    • We compile fondant pipelines to the IR YAML format, which is supported by other execution engines such as Vertex
    • You need a KfP v2 cluster to run fondant pipelines

Fixes

  • Fix data explorer for usage on Windows
  • Fix propagation of client_kwargs argument to configure Dask Client

Components

  • Every reusable component now has a clear README describing its usage
  • Add load_from_parquet component to load parquet files as input data
  • Add embed_text component to embed documents and other text
  • Add chunk_text component to chunk documents into passages
  • Add index_weaviate component to index data in a weaviate vector store
  • Fix issue with mixed type ids in LAION retrieval components
  • Improve success rate of download_images component
  • Fix OOM issues for inference components using GPU
  • Limit data read by load_from_hub component to used columns

Detailed changes

New Contributors

Full Changelog: 0.5.0...0.6.1