Label Studio is an open source data labeling tool that supports multiple projects, users and data types in one platform. Label Studio can perform different types of labeling with many data formats and integrate with M/L backends.
I am very excited to see LabelStudio reach their 1.8 milestone. I think having high quality open source tools that helps make us humans more effective at fine-tuning LLMs like LLaMA and other foundation models is important as we pull our data in to our enterprise specific AI brains.
@irina_haupt Good question, Irina! Label Studio is a data labeling platform that allows data scientists to fine-tune LLMs through human-in-the-loop-style retraining or RLHF retraining.
To put it simply — this is the "interface" where one can provide feedback to the model and coach it to better fit your needs.
You can read more about model retraining with Label Studio here: https://labelstud.io/blog/human-...
or check out a demo on our GitHub here: https://github.com/heartexlabs/RLHF
Hey Product Hunt! Generative AI and large language models like ChatGPT and LLaMa have brought innovations in machine learning and new MLOps processes. The need for enhanced tooling to leverage these models to build domain-specific applications is growing rapidly.
That's why, with our 1.8.0 release of Label Studio, we’ve made the process of fine-tuning LLMs and foundation models even easier. 💪🏻🚀
Some of the new features we're excited about in this release Include:
🆕 Ranker Interface
Label Studio now has a new, improved interface designed for ranked choice model retraining. This Ranker interface is designed to work with ranking models that predict a relevance score. This brings an updated tag that can be configured to evaluate and compare responses.
🆕 Generative AI Templates
We’re introducing new labeling templates specifically designed for generative AI! Check out the templates for:
* Supervised language model fine-tuning
* Human preference collecting for RLHF
* Chatbot model assessment
* LLM Ranker
* Visual ranker
🚀 Interface Improvements
We've added side panel updates to better display important details about what you’re annotating, including relations, comments, and history.
There's a new annotations tab carousel that allows you to view all annotations simultaneously!
We've relocated the control button panel for an improved workflow.
Thanks for checking us out!
@nate_kartchner features look incredibly promising. The advancements in generative AI and large language models are transforming the way we handle machine learning tasks, and Label Studio seems to be keeping pace with these advancements brilliantly.
The new Ranker Interface stands out as a significant upgrade, promising a more intuitive and effective way to retrain ranking models. The potential it offers in terms of evaluating and comparing responses is something I'm looking forward to exploring.
@sarah_harris5 Thanks so much! It's pretty exciting to see the ways in which foundation models are changing the way machine learning operations are run. Human feedback will always have value, but we've already seen the ways in which it's valuable to machine learning change, and I expect it will continue to do so. Our goal is to keep developing to make sure that we keep up with that. :)
Enjoy your explorations with Ranker - we'd love to get your feedback as you do so!
Marketing Machines
Collato
Label Studio
Collato