February 2024 Community News!

🚀 Tutorial: Integrate LLMs and Prompt Interface to Automate Data Labeling

Learn how integrating Large Language Models (LLMs) into the data labeling process can revolutionize dataset curation. While our previous articles focused on using LLMs for context-aware predictions, this blog delves into a prompt-centric workflow that enables real-time prompt engineering and leverages LLMs as collaborative partners in the annotation process. This dynamic approach not only streamlines the annotation workflow but also continually enhances the quality of the dataset and prompt.

Follow along with this practical tutorial to classify chatbot intent by integrating ChatGPT-4 and the interactive prompt interface into Label Studio, including example notebooks and code.

🎥 Video tutorial: labeling podcast transcripts to fine-tune an ML model

Label Studio user Mike Heavers, Staff Design Technologist on Mozilla's Innovation Team, created a step-by-step video explainer using Label Studio to fine-tune a model to better understand the nature of podcasts.

Mike previously used the technique of Retrieval-Augmented Generation (RAG) to pass podcast transcripts on the fly to a ChatGPT model, but the model still didn’t perform well enough when differentiating podcast elements or recognizing which content was most important. For example, it was typically pulling content from the beginning or end of the podcast (which highlighted sponsors) rather than pulling from the middle, where all of the good stories and conversations take place.

By converting and labeling podcast transcripts, Mike created a high quality dataset to improve the accuracy of the model.

🐿️ More Resources

📝 We've made updates to the Label Studio docs! The create project page has been updated, and we've added a new project settings page to address frequently asked questions. The create regions without adding labels instructions have also been updated based on community feedback in Slack. 

❓From the community slack archives: Are you getting a runtime error while trying to export thousands of entires? If the export times out, see how to export snapshots using the SDK or API. You can also use a console command to export your project.

🚆 Pulumi, an open source tool for deploying and managing cloud infrastructure, published a resource about deploying the Label Studio helm chart on AWS EKS using Pulumi packages. 

🤸 Thank you for being part of the community!

Do you know of a game-changing ML integration that's improved your labeling workflow? Do you have your own Label Studio tips and tricks to share? Head on over to the Label Studio Slack Community!

Happy Labeling!