Virtualized Elastic KV Cache for Dynamic GPU Sharing and Beyond

Boosting GPU utilization for LLM serving via dynamic spatial-temporal prefill & decode orchestration

TensorFusion landing page and product docs

Tricks for GPU sharing pods on Kubernetes without any use of middleware like HAMi or DRA

SLURM-native software GPU slicing for NVIDIA clusters using memory limits and compute time-slicing.

GPUs unite using secure and private crypto transactions to distribute compute to decentralized nodes.

Distributed peer-to-peer LLM inference network. Volunteer your GPU, earn AI credits, run any open-source model for free. Anonymous, encrypted, unstoppable.