XARIV

XARIV Relay

Open weights on your machine — not our GPUs

Download a GGUF, deploy llama.cpp on localhost, then compare two candidates side by side. Vercel cannot host model files or llama-server, so the public site ships a timing replay for demos.

Compare F16 vs Q4Kernel fusion replayLive local deploy
  • Website (this page): split-pane speed demo, no download.
  • Your Mac: real weights cache in ~/.xariv/relay/models/ after the first Hugging Face fetch — Deploy does not re-download if the file is already there.

Docs →