We’re excited to announce the latest updates across our platform, focused on boosting performance, expanding model support, and enhancing retrieval-augmented generation (RAG) capabilities. This release brings meaningful improvements to Composer, Layar API, and Curate—making your workflows faster, smarter, and more reliable.
🔍 Product Update Highlights
Composer: Support for new models, faster tokenization, and improved vector database performance.
Layar API: Enhanced RAG performance, streaming support, and better throttling.
Curate: Multi-label classification with new visualization tools.
📦 Composer
Model Enhancements: Added support for DeepSeek, Blossom, LLama4 models.
Tokenizer Improvements: Introduced caching for tokenizers at startup to reduce request latency.
Error Handling: General improvements to error message details.
Infrastructure: Added scripts for vector database schema migration.
Vector Database improvements: Schema updates, deterministic UUIDs, and memory optimizations, streamline validation checks during generation to improve throughput
RAG Enhancements: Improved table formatting when sending tabular data to a prompt for RAG
🧠 Layar API
Caching: Implemented caching for RAG document sets for faster retrieval
Embeddings: Switched to distributed search command.
Streaming: Enabled streaming for OAI endpoint.
Throttling: Introduced LLM request throttling from Curate.
Bug Fixes: Addressed issues with data fabric, tag handling, and document deletion.
Enhancements: Added support for downloadable models
Data Source Update: Updated importer in clinical trials data source.
Featurizers: Added configurable featurizers and improved automated testing framework for Chemistry prediction jobs
Bug Fix: Improved logic for determining the last time a user was active in the system
🧬 Curate
Multi-label Support: Added toggle and confusion matrix visualization.