We're thrilled to unveil Milvus 2.4, a milestone in our quest to enhance search capabilities for vast datasets. This version accelerates search efficiency and broadens our horizons toward a unified search platform capable of fulfilling diverse search use cases with unparalleled speed and precision. Our efforts to support accelerated and unified search reflect our dedication to delivering a robust, scalable vector data management and querying solution.
📚 Docs: https://milvus.io/docs
🛠️ Release Notes: https://milvus.io/docs/release_notes.md#240
🐳 Docker Image: docker pull
🚀 Release: Milvus-2.4.0
Key Highlights
🚀 CAGRA Index Support — Milvus 2.4 introduces support for NVIDIA’s CAGRA Index, setting a new standard in GPU-accelerated graph indexing for search applications requiring extremely low latency. Outperforming the industry standard HNSW index, CAGRA demonstrates exceptional efficiency and speed, even for small batch sizes. Its advantages become even more pronounced with larger batches, catering to the requirements of high-throughput environments. Additionally, Milvus incorporates brute-force search options with CAGRA for use cases where achieving the highest possible recall is essential. Developers can explore the documentation to start leveraging CAGRA GPU index capabilities.
Milvus raft cagra vs milvus hnsw
🔍 Multi-vector Search — The addition of Multi-vector Search in Milvus 2.4 enhances AI application development by enabling the management of multiple vector searches and reranking within its framework. This simplifies handling multimodal searches and improves the recall of retrieving information. The ability to store and query multiple vectors for a single entity within a collection makes it easier to model data in a way that's natural and efficient for real-world applications. This feature also makes integrating and optimizing custom reranked models straightforward, supporting the development of advanced search functionalities, like accurate recommender systems that leverage insights from multi-dimensional data.
How the multi-vector search feature works
🧮 Grouping Search — Another highlight of Milvus 2.4 is the Grouping Search feature, enhancing efficiency in compute resources and developer productivity for grouped search queries. This feature addresses the challenges of querying large datasets, like documents or videos segmented into vectorized chunks or frames. Previously, aggregating search results by attributes was cumbersome, requiring extensive compute resources and complex coding workarounds. Now, Grouping search allows for straightforward aggregation, enabling users to retrieve top results by the grouped fields directly.
Example- search books contain "harry potter"
🔮 Sparse Vector Support (beta) — With much anticipation, beta support for sparse vector embeddings is available in Milvus 2.4, enabling efficient, semantically rich ANN searches. This feature is designed for embeddings from neural models like SPLADEv2 and statistical models such as BM25, enhancing search capabilities beyond traditional keyword searches by focusing on semantic similarity. Slated for general availability later, this beta phase aims to optimize performance and integration within Milvus's ecosystem. It opens new avenues for hybrid search—merging keyword and embedding strategies—to boost text search accuracy significantly. Ideal for users transitioning from keyword-based systems and seeking enhanced search accuracy without the need for deep customization, sparse vector support moves Milvus towards more nuanced, hybrid text search methodologies.
⬆️ Other Key Enhancements — Milvus 2.4 brings many other new features and enhancements, including Regular Expression support for enhanced substring matching in metadata filtering, a new scalar inverted index for more efficient scalar data type filtering (thanks to Tantivy), and a Change Data Capture tool to monitor and replicate changes in Milvus collections. Together, these updates significantly boost Milvus's performance and versatility for complex data operations.
Getting Started
Interested in exploring Milvus 2.4.0? Start with the release notes for an overview of the new features and improvements. Join our upcoming webinar with James Luan, Zilliz’s VP of Engineering, for an in-depth discussion on Milvus 2.4.0's capabilities. Ready to see the changes firsthand? Download Milvus 2.4.0 and begin integrating these enhancements into your projects.
Start Free, Scale Easily
Try the fully-managed vector database built for your GenAI applications.
Try Zilliz Cloud for Free