Thank you for being a valued Indico user. The Indico Data team is excited to launch 5.4, the newest addition to our continually improving platform. These release notes are home to all release documentation regarding 5.4, including our patch release notes.

Innovations and Updates in v5.4:

• Find the dataset you need quickly with type-ahead search bars
  • During workflow creation, when selecting your dataset, type the name of your dataset into the search bar
  • From all of your datasets, your list will narrow down to help you quickly and easily find the dataset you’re looking for
•  Learn more about your models in Explain
  • Look to the header on your Explain details page to find information like when a model finished training and the number of examples used
• Streamline your access to models
  • When troubleshooting, don’t spend your important time worried about getting model access
  • Now, by default, users with manager or analysts roles on a dataset have access to all models and teach tasks associated with that dataset
• Submit more files to your workflow at once
  • Submit up to 300 files at once from your Workflows canvas
• The submissions list page keeps getting better!
  • Deleted documents are now denoted on the submissions list page
  • This update has Submissions list API updates too! You can now filter your submissions by:
    • created at
    • updated at
    • has files
    • and file type
• Document reports now include each submission’s file type
• Get notified when someone else is reviewing the same document
  • When reviewing a document, if another user is currently reviewing the same document, you will now receive a message prompting you to return to the submissions list

Patch Releases for This Version:

• 5.4.1
  • This patch contains CVE fixes
• 5.4.2
  • This patch corrects a performance regression that is exacerbated by high load on the cluster for extended periods of time, eventually resulting in task failures
• 5.4.3
  • In specific circumstances, extraction model training was taking longer than normal. This patch contains the fix.
• 5.4.4
  • This patch contains fixes for URL retrieval errors users were receiving under certain circumstances.
  • In order to improve worker creation, this patch updates some object detection variables.
  • In certain situations, examples with labels that contained large numbers of start/end instances were improperly loading. This has been corrected.
• 5.4.5
  • Previously, adding additional data to complex workflows would result in errors. This has been corrected.
  • In certain circumstances, confidence scores have been updated in this aptch to account for negative confidences.
• 5.4.6
  • Under certain circumstances, specific kinds of models were unable to calculate metrics. This has been corrected.
• 5.4.7
  • This patch allows administrators to disable the widget in Explain that retrieves extraction examples and may be causing their instance to slow down.
• 5.4.8
  • Default limits for gather_with_concurrency have been updated to optimize for speed.
  • This patch contains a fix that allows models to complete training even when unable to calculate metrics.