Runs beneath unmodified vLLM and lets each GPU serve many models while holding first-token and per-token deadlines. Measured 10 models per GPU, no answers lost.
Projects
What ZenEdgeX is building. Each project has its own pages, its evidence, and what is measured, estimated or still planned.
Runs an unmodified vision-language-action robot policy faster and proves, tick by tick, that the actions are bit-identical to the policy as trained.
AI-powered chip design exploration. Design better chips, faster, with architecture and compiler optimized together for your workloads.
A digital in-memory-compute AI accelerator IP for language and vision models at the edge, with a RISC-V control core, generated and verified for your model.