PureKVA v1.2 now GA
Pure KVA (Key-Value Accelerator) is an enterprise AI solution designed to speed up large language model (LLM) inference by storing and reusing precomputed attention key-value states.
Want to learn more about PureKVA and what it can do for you? Check out the link here for more information and how to get started:
https://support.everpuredata.com/r/pure-kva/pure-kva-overview-and-architecture
Version 1.2 is now available for download. Check out the highlights of what's new below.
What is new in v1.2
- PureKVA v1.2 adds the following capabilities:
- Containerized deployment in Openshift/Kubernetes
- S3 and S3overRDMA support
- Multi-tenancy with storage isolation support
- Three-tiered memory architecture for improved performance
- Nemotron model support
- PureKVA now supports offloading TurboQuant-quantized KV caches for vLLM.
Get v1.2:
https://support.everpuredata.com/r/pure-kva/pure-kva-release-v1-2
