Making the RK3588 Mali-G610 GPU a real LLM prefill accelerator via the open Vulkan stack (panthor + Mesa panvk + llama.cpp). Two upstreamable fixes -> 4.8x prefill, no vendor blobs.
Updated 2026-07-12 05:05:05 +00:00
Kernel source/branch/patch curation, per-host build orchestration, promote-to-fleet pipeline for the home fleet. Peer agent to His (home infra).
Updated 2026-05-19 21:44:31 +00:00