Skip to content

Qualify Soprano and LFM3 on hardware outside the v0.8.0 pass (CUDA, Windows, iOS, Intel Mac) #147

Open

Description

@leehack

Soprano (#123) and LFM3 (#130) pass on the published v0.8.0 bundles on macOS arm64 (M4 Max), GCE Linux x64 and arm64 (CPU), and Android CPU on three phones. Mali-G715 and Xclipse 530 Vulkan pass; Adreno 750 Vulkan fails under #79. Evidence: llamadart-evidence/v080-qualification/.

Not run, so no support claim:

  • NVIDIA CUDA on Linux and Windows (needs a GPU instance; not started, per the no-GPU scope of this pass)
  • Linux Vulkan, HIP and BLAS on real devices
  • Windows x64 and arm64 CPU (arm64 hardware: #65)
  • iOS device and macOS x86_64; Intel/AMD Metal stays with #116
  • Android x64: Test Lab offers no x86_64 device
  • Omni and d1-Omni on Android Vulkan, d1-Omni on Android at all (the phones ran d1-2B only)

For each: qual and score_audio.py from the evidence kit, with real output checks as on the qualified targets.

Activity

  1. leehack commented on Oct 30, 2026

    @leehack
    OwnerAuthor

    Triage: P3 enhancement, qualification only. No release scope; run a target when core 0.18 is about to claim it. CUDA needs a GPU instance, which needs the maintainer's go-ahead.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    enhancementNew feature or requestpriority:P3Useful cleanup or longer-term work

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions