From Dev to Prod, Vector Database Made Easy
0 likes185 views
Zilliz, the company behind the Milvus project, specializes in vector databases that store embedding vectors for semantic retrieval of unstructured data. Milvus, an open-source project since 2019, aims to simplify the management and deployment of vector databases with scalability, performance, and cost-efficiency at its core. Various deployment options are available, including self-managed and fully-managed cloud services, targeting companies at different growth stages in their AI development lifecycle.
From Dev to Prod, Vector Database Made Easy
- 1. 1 | © Copyright 10/22/23 Zilliz 1 | © Copyright 10/22/23 Zilliz From Dev to Prod, Vector Database Made Easy Presented by: Charles Xie
- 2. Charles Xie Founder and CEO of Zilliz, Founder of the Milvus Project ● Board member of LF AI & Data, Foundation and chairperson from 2020 to 2021 ● Founding engineer of Oracle 12c cloud database ● Master in Computer Science, University of Wisconsin-Madison 2 ● means zillions of zillions, pronunciate as /ʼzilis/ ● the company behind the Milvus project, since 2018 ● Series B, $113M funding
- 3. 3 Vector Database : making sense of unstructured data 2024 A vector database stores embedding vectors and allows for semantic retrieval of various types of unstructured data.
- 4. 4 Milvus, OSS vector database since 2019 Originally created by Zilliz, hosted by the Linux Foundation 2024 28K 10000 GitHub Stars Enterprise users 70M Downloads 270 Contributors
- 5. Vector database made easy for companies at different stages
- 6. GenAI company life cycle
- 7. Bootstrapping stage ● First to market ● Easy to use ● Prepare for future growth
- 8. 8 2024 Easy to start ● Pip-install on your laptop ● plug into your favorite AI dev tools ● push to production with a single line of code
- 9. 9 2024 Prepare for future growth ● Write your code once, and running everywhere, at any scale ○ API and SDK are all the same Milvus Lite ● Embedded ● No server installation ● Low footprint Milvus ● Dedicated server ● High performance ● Easy to maintain ○ K8S ○ Docker Zilliz Cloud ● Multi-tenancy ● High availability ● Data security ● Complianced ○ SoC2 ○ GDPR
- 10. Growth stage ● ROI, Cost optimization ● Scalability
- 11. 11 2024 Status quo: expensive, not scalable In memory ● Most of other VDBs ● HNSW ● Expensive
- 12. 12 2024 Lower cost In memory ● Most of other VDBs ● HNSW ● Expensive On Disk ● Up to 10x lower cost
- 13. 13 2024 Lower cost In memory ● Most VDBs ● HNSW ● Expensive On Disk ● Up to 10x lower cost Object store ● Up to 50x lower cost
- 14. Storage and computation separation
- 15. Milvus: decoupling computation and storage
- 16. 16 | © Copyright 10/22/23 Zilliz 16 | © Copyright 10/22/23 Zilliz hierarchical storage and caching
- 17. 17 | © Copyright 10/22/23 Zilliz 17 | © Copyright 10/22/23 Zilliz Data caching
- 18. 18 2024 Higher scalability 10B vectors of 1536 dimensions in a single Milvus/Zilliz Cloud instance
- 19. 19 2024 Higher scalability 10B vectors of 1536 dimensions in a single Milvus/Zilliz Cloud instance 100B vectors in one of the largest deployment
- 21. Expansion stage ● Performance ● Avoid vendor lock in ○ Move data when you want ● Multi-cloud ● Global availability
- 22. 22 2024 VectorDBBench : OSS framework for VDB benchmarking https://github.com/zilliztech/VectorDBBench
- 24. 24 Multi-cloud: Zilliz Cloud is built atop of OSS Milvus AWS, GCP, Azure 2024
- 25. Global availability: Zilliz Cloud has 20 availability zones NA, EMEA, APAC
- 27. 27 Retrieval-Augmented Generation RAG 2024 A technique that combines the strength of retrieval-based and generative models: ● Improve accuracy and relevance ● Eliminate hallucination ● Provide domain-specific knowledge
- 28. 28 RAG : an economic perspective 2024 A business model that bridges public data and private data ● Data sovereignty ● You can't and shouldn't give your private data to others
- 29. 29 RAG Evolution 2024 RAG 1.0, last year ● text ● LLMs, ○ GTP3.5, GPT4
- 30. 30 RAG Evolution 2024 RAG 1.0, last year ● text ● LLMs, ○ GTP3.5, GPT4 RAG 2.0, this year ● image, video ● multi modality models ○ GTP4o
- 31. 31 RAG Evolution 2024 RAG 1.0, last year ● text ● LLMs, ○ GTP3.5, GPT4 RAG 2.0, this year ● image, video ● multi modality models ○ GTP4o RAG 3.0, next year? ● user behavior ● customized recommendation systems ○ Merlin
- 32. More application scenarios of vector databases
- 33. Thank You ! Milvus Open Source Self-Managed github.com/milvus-io/milvus Zilliz Cloud SaaS Fully-Managed zilliz.com/cloud