acoustic-carpenter-78188
04/21/2026, 12:24 AM--gpu to flyte start demo. When set, the underlying docker run is invoked with --gpus all, giving the demo container access to host NVIDIA GPUs.
flyte start demo --gpu --image ghcr.io/flyteorg/flyte-demo:gpu-latest
Default off — existing non-GPU users are unaffected.
Why
The GPU-capable demo image being added in flyteorg/flyte#7243 configures k3s, containerd, and the NVIDIA device plugin inside the image, but none of that matters if the host GPUs aren't passed through to the container. --gpus all on the docker run is the piece the CLI owns.
Change
• _start.py — new --gpu click option on the demo command.
• _demo.py — threads gpu: bool through launch_demo → _launch_demo_rich/plain → _run_step → _run_container, which appends --gpus all when true.
Test plan
Screenshot 2026-04-21 at 12 38 07 AM
• Installed locally with `uv tool install --from . flyte`; flyte start demo --help lists the new flag.
• flyte start demo --gpu --image flyte-demo:gpu-local on an A10G host launched the container with --gpus all, a Flyte task with Resources(gpu=1) ran torch.cuda.is_available() == True.
• No regression expected on existing users (flag defaults to off).
flyteorg/flyte-sdkacoustic-carpenter-78188
04/21/2026, 4:45 PM