W&B Weave updates: New display names, chat tab, and events galore
We're constantly improving W&B Weave. Here's what we've been up to lately.
Created on October 10|Last edited on October 10
Comment
Welcome to the first W&B Weave newsletter of October. This week, we’re introducing custom display names and a new chat tab in W&B Weave, plus a host of hands-on events to get you building with LLMs faster.
But as always, let’s start with our tip of the week:
LLM tip of the week ✅
Query refinement for RAG is like giving your system X-ray vision. The system can “see“ user intentions more clearly, leading to increased retrieval accuracy and higher quality LLM responses.
Our latest on-demand course, RAG++: From POC to Production, walks you through four key query refinement approaches as well as evaluation for RAG systems, advanced retrieval and re-ranking, and much more. Try it now and get free API credits to get your started.

Product news 🚀
Control the display names of calls in W&B Weave
Keep your work organized by setting the display name of calls based on inputs, outputs or any other custom logic.

Chat tab
See the entire conversation history quickly with the new Chat tab. Easily browse to find past messages, tool use, and context efficiently. Chat details are automatically displayed when selecting calls that conform to popular LLM chat message formats. We're continuing to add support for more formats, so please reach out if this isn't displaying for a certain LLM provider.

Popular blogs 📑
Llama 3.2 for multi-modal RAG over financial filings
Often documents contain images with important information that you want to be made available to your RAG system. One approach is to use a VLM like Llama 3.2 90B to extract the text and descriptions from the images for later retrieval. We’ll walk you through how to build your own pipeline, including the code to get you up and running.

HEMM leaderboard: Holistic evaluation of multi-modal generative AI models
The HEMM leaderboard is a collection of comprehensive evaluations of text-to-image generation models for prompt comprehension. Each model is evaluated on a set of 716 prompts with complex actions and interactions between objects such as "the rectangular mirror was hung above the marble sink.” The generated images are then evaluated using a VLM with Flux 1.1 Pro currently leading the rankings.
o1 quickstart
Getting started running the new o1 models from OpenAI via the API is surprisingly easy and will offer much deeper reasoning capabilities for more complex tasks. In this quickstart using o1-preview, we'll have you ready to go in about 5 minutes.
Events 🏢
🇫🇷 AWS GenAI Loft Paris - October 14th at 15:00 in Paris
Come join a session on building production-grade RAG powered applications with Weights & Biases Weave and AWS. Specifically, we’ll showcase how to use W&B Weave with Amazon Bedrock.
🌉 Build a production-ready LLM app in 90 minutes with W&B Weave - October 16th at 6pm PT in SF
Join us for a 90 minute hands-on workshop where you'll set up everything you need for a reliable generative AI product at the AWS GenAI Loft.
🏙️ Optimizing AI evaluations - October 16th at 6pm ET in NYC
In partnership with Google, we will host sessions covering productionizing LLM-as-a-judge systems, aligning them with human judgments, and building annotation UIs.
🌉 GenAI salon - October 17th at 5pm PT in SF
Join us in person in our San Francisco office. Jerry Liu (CEO of LlamaIndex) and Ben Firshman (Founder & CEO at Replicate) will explain how they built LlamaIndex and Replicate.
Community
This project focuses on evaluating various LLMs, specifically different implementations of LLaMA 3.1 70B, to determine their effectiveness in understanding and generating dental knowledge. It's a fun and intriguing exploration into the intersection of artificial intelligence and dentistry.
Slide doctor was built during the MistralAI October 2024 hackathon and transforms your PowerPoint presentations into polished, professional masterpieces using AI. Complete with a Gradio app, it generates better slide layouts—plus, it catches spelling mistakes.
Need help getting started with W&B Weave?
Add a comment
Iterate on AI agents and models faster. Try Weights & Biases today.