• build-11-0630a52 0630a52354

    build-11-0630a52
    All checks were successful
    Build and Release / build (push) Successful in 13m32s
    Stable

    taxmalalas0001 released this 2026-10-17 19:51:20 +00:00 | 17 commits to master since this release

    Commit: 0630a52354

    Build: release

    Message:

    Run the on-device model on the GPU, falling back to CPU

    LlmInferenceEngine tries Backend.GPU() first and falls back to the CPU
    when GPU init fails. A GPU driver can also kill the process natively,
    which no catch sees, so an in-progress flag is written before the attempt
    and cleared after; if a launch finds it still set, that device is pinned
    to CPU from then on. Declare the vendor OpenCL libraries with
    uses-native-library (required=false) so Android 12+ lets LiteRT load them.

    Downloads