Skip to content

install.sh: detect platform, install the Linux/arm64 CUDA bundle - #33

Open
prabod wants to merge 1 commit into
mainfrom
feat/install-linux-cuda
Open

install.sh: detect platform, install the Linux/arm64 CUDA bundle#33
prabod wants to merge 1 commit into
mainfrom
feat/install-linux-cuda

Conversation

@prabod

@prabod prabod commented Jul 26, 2026

Copy link
Copy Markdown
Contributor

The installer was hard-locked to macOS/Apple-Silicon (err "BaseRT requires macOS"), so CUDA/GB10 users had no curl -LsSf … | sh path.

Changes

  • Platform detection picks the matching prebuilt bundle:
    • Darwin/arm64basert-engine-macos-arm64 (unchanged)
    • Linux/aarch64 + an NVIDIA GPU/driver → basert-engine-linux-arm64-cuda (new, published by the baseRT release workflow)
    • anything else → a clear error.
  • Clean up libbaseRT.so* on upgrade (not just the macOS .dylib).
  • Point the "try it" example at the hosted catalog id basecompute/Qwen3-0.6B (resolves through the hub, backend-aware) instead of a raw Qwen/… HF repo.

The installed CLI's client_backend() resolves to cuda on the Linux build, so basert pull fetches the CUDA catalog variant automatically.

The installer was hard-locked to macOS/Apple-Silicon and errored out on Linux,
so CUDA (GB10) users had no `curl | sh` path. Detect the platform and pick the
matching prebuilt bundle: macOS/arm64 → basert-engine-macos-arm64, Linux/arm64
with an NVIDIA GPU → basert-engine-linux-arm64-cuda (new; published by the
baseRT release workflow). The installed CLI is built for that backend, so
`basert pull` resolves the matching catalog variant automatically.

Also: clean up libbaseRT.so on upgrade (not just the macOS .dylib), and point
the "try it" example at the hosted catalog id (basecompute/Qwen3-0.6B) instead
of a raw HF repo.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant