We are delighted to announce the latest release of Snorkel Flow. Inspired by user feedback, the 2023.R3 Release is packed with improvements that amplify user experience, streamline workflows, and enhance performance, ensuring our users derive unparalleled value from our platform.

Image6

A deep dive into the new streamlined app creation workflow

We’ve redesigned our app setup workflow to allow users to effortlessly set up new AI applications on Snorkel Flow with just a few clicks. This new guided workflow is designed to ensure success for your AI use case, regardless of complexity, catering to both seasoned data scientists and those just beginning their journey.

Snorkel Flow’s new onboarding workflow streamlines the AI development process by guiding you through connecting your data, setting application parameters, and selecting the right preprocessors—all from a single access point. While creating your app, you’ll receive a preview of your dataset, allowing you to identify and correct critical data errors early.

Image5

Enhanced user experience in Snorkel Flow Studio

We’ve made significant improvements to Snorkel Flow Studio, making it easier for you to export training datasets in the UI, improving default display settings, adding per-class filtering and analysis, and several other great enhancements for easier integration with larger ML pipelines. Large teams collaborating on Snorkel Flow will also enjoy our new comment-based filtering making it easier to communicate with teammates and more easily address outliers to ensure the highest quality data possible.

Image2

Expanded FM access via Google PaLM API integration 

We’re pleased to announce that we’ve expanded our foundation model (FM) library to include PaLM 2 and other models in the Vertex AI Model Garden. Users can access these new models from a pull-down in the Snorkel Flow interface. 

After accessing unstructured data in Google Cloud with a few clicks via the native BigQuery connector, machine learning teams can select PaLM 2 and evaluate zero-shot results. It’s easy to apply and evaluate advanced prompting techniques within Snorkel Flow, then use guided error analysis tools to identify specific areas where data is weak and collaborate with internal subject matter experts to fine-tune Google models with proprietary data and then distill results from multi-billion parameter models to a much more efficient specialized model.
Snorkel Flow integration with Vertex AI makes it easy to plug into established MLOps pipelines and deploy a fine-tuned or specialized model via Vertex Endpoints. When Vertex Model Monitoring detects data drift,  input feature values are submitted to Snorkel Flow, enabling ML teams to adapt labeling functions quickly, retrain the model, and then deploy the new model with Vertex AI.

Image4

Revamped Snorkel Flow SDK 

Also included in the 2023.R3 Snorkel Flow release is an upgraded Python SDK, now enhanced with advanced data preparation capabilities that enable on-the-fly transformations. This allows for fine-grained control, making it easier to spot and resolve inconsistencies at crucial points in your data-centric AI development workflow—no need to start the your process over if you need to make updates to your dataset. Along with these new features, you’ll discover documentation that’s not only clearer but also more intuitively organized for smoother navigation.

Image1

Multi-label & high-cardinality text classification just got faster

Performance enhancements are the heartbeat of every Snorkel Flow release. But we took things a step further with this release to improve multi-label text classification and ML tasks with more than 100 labels.

  • A 4x speedup on long-running training processes.
  • Up to 10x speedups on key operations for more interactivity.
  • A revamped user interface that prioritizes legibility and usability, with special attention given to multi-label interactions.
Image3

That’s a wrap on the 2023.R3 release, and we’ll be back soon with 2023.R4. Until then, keep building better AI faster with Snorkel Flow.

Learn how to get more value from your PDF documents!

Transforming unstructured data such as text and documents into structured data is crucial for enterprise AI development. On December 17, we’ll hold a webinar that explains how to capture SME domain knowledge and use it to automate and scale PDF classification and information extraction tasks.

Sign up here!