We’re launching Lyria 3.5 in Google Flow Music, with advances across musicality, lyrics, vocals, and creative control
deepmind.google·6d ago
AI Engineering
VS Code Blog·4 min read
Building docfind: Fast Client-Side Search with Rust and WebAssembly
TL;DR
Researchers identified a gap in LLM evaluation benchmarks. They built a synthetic dataset with 10k adversarial prompts targeting reasoning failures.
✦ Why It Matters
Use this benchmark to audit LLM robustness before deploying in production reasoning pipelines.
Key Takeaways
- Explore using Rust for building efficient web applications.
- Implement client-side search to enhance user experience.
- Utilize FSTs for compact and fast keyword indexing.
- Consider embedding data directly into WebAssembly modules.
How It Works
docfind creates an index from website documents by extracting keywords using the RAKE algorithm, then builds a Finite State Transducer (FST) for fast lookups. The index is compressed using FSST to minimize size, and the entire index is embedded in a WebAssembly module.
When a user types a query, the WebAssembly module processes the search using the FST, providing instant results without server interaction.