September 2022 Summaries

Vector databases like Weaviate use Machine Learning models to offer various functionalities such as summarizing text, auto-labeling data tokens, converting data into vectors for context-based queries, etc. However, running model inference in production is challenging due to the need for expensive specialized hardware and underutilization of resources. Companies like Hugging Face, OpenAI, and Cohere offer running model inference as a service. Weaviate v1.15 introduces support for Hugging Face Inference API, allowing users to use pre-trained sentence similarity models or upload their own private models. The integration is fully automated and optimized, with the option to choose between text2vec-huggingface and text2vec-openai modules.

Weaviate v1.15.1 is a significant patch release addressing over 15 bugs in indexing, sorting, aggregation, stability, and UX. The release includes fixes for vector and object/inverted stores, improved sorting implementation with new tests, seven aggregation issues fixed, a rare critical bug related to data race, and several UX improvements. This release ensures increased stability, proper updates reflection, and better user experience.

Vector search is incredibly fast due to its use of Approximate Nearest Neighbor (ANN) algorithms, which trade off a bit of accuracy for significant speed gains. These algorithms maintain good performance on large-scale datasets and allow for efficient vector searches. Weaviate, a vector database, uses ANN algorithms like HNSW to offer ultra-fast queries while maintaining high recall rates, throughput, and low latency.

Weaviate 1.15 has been released with new features, performance improvements, distance metrics, and modules. Key updates include cloud-native backups for selected classes or the whole database, reduced memory usage by up to 30%, better control over garbage collector through GOMEMLIMIT, faster imports for ordered data, more efficient filtered aggregations, two new distance metrics (Hamming and Manhattan), and two new Weaviate modules (Summarization and HuggingFace).

Weaviate, a leading open-source vector database technology provider, has introduced a new pricing model for its Weaviate Cloud service. Inspired by "pay-as-you-grow" cloud storage pricing, users can now pay a usage-based rate for search on a per-dimension basis, starting at $0.05 per million dimensions. This makes it easier and more affordable for enterprises of any scale to utilize Weaviate's vector database technology. The new model aligns with the company's commitment to creating truly open-source software, as customers using the free service will always have access to all of Weaviate's vector database capabilities.