Navigating Vector Databases: Indexing Strategies, GPU, and More

On this episode of the ML Platform Podcast, Frank Liu discusses the basics, problems and challenges of vector databases, including indexing strategies, segmentation, vector lengths used in production, GPU-accelerated vector databases, potential use cases, and more.
Timestamps:
00:00 Introduction
01:18 Who is Frank?
02:18 Vector databases 101
11:49 Embedding vector lengths used in production
13:27 Indexing strategies for vector databases
20:57 Vector updates and segmentation
27:43 The problem of updating embeddings and re-indexing
33:37 Milvus Lite, Milvus Standalone, and Milvus Cluster
35:09 GPU-accelerated vector databases
40:18 When to consider adding a vector database to your tech stack
47:44 Combining filtering with vector search
01:00:00 Combining keyword and vector search
01:04:36 Building a documentation chatbot with a vector database
01:09:14 Closing remarks
Resources:
ImageBind: Holistic AI Learning Across Six Modalities: ai.meta.com/blog/imagebind-si...
Frank’s website: frankliucs.com/
Milvus: milvus.io/
OSS Chat: osschat.io/
Follow us & stay updated:
► Vist our website: neptune.ai/?...
► Follow us on Linkedin: / neptuneai
► Follow us on Twitter/X: / neptune_ai
► Check our Github: github.com/neptune-ai
Connect with Piotr on Linkedin: / piotrniedzwiedz
Connect with Aurimas on Linkedin: / aurimas-griciunas
Connect with Frank on Linkedin: / fzliu
Connect with Frank on Twitter/X: / frankzliu
The episode was recorded on 7 September 2023, and some information may not be up-to-date.
#vectordatabases #milvus #mlops #ml #llms