Skip to content

first-run: CPU-only hosts seed no Main slot (derive_profile cpu→vulkan incoherent) #834

@thinmintdev

Description

@thinmintdev

Follow-up from #833 (hal0 setup TUI).

On a GPU-less host, hal0 setup --auto derives device=cpu then derive_profile("chat","cpu")→"vulkan". SlotManager.create()'s #807 coherence check rejects device=cpu + profile=vulkan, so no Main slot is seeded (apply_setup catches it best-effort, sentinel still written). The old install.sh probe block had a proper CPU fallback (backend=cpu).

Fix: add a CPU-coherent llm profile to SEED_PROFILES and map derive_profile(chat/coder, cpu) to it (instead of vulkan). Affects src/hal0/install/profile_derive.py + the profile seed set. Not blocking for Strix Halo (derives gpu-rocm/rocm-mtp correctly).

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type
    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions