Steerling-8B on AnyCloud

Run Guide Labs' public Steerling-8B generation path as a finite, portable GPU job:

anycloud job ghcr.io/anycloud-sh/steerling-8b:0.2.0-f2e5952-r2 --gpu-type a10 --gpus all

The immutable tag resolves to sha256:f2e38b4fc240a4134f54a6c877db306411d3df4a7105a06c933cb5de5659de8f.

This is an AnyCloud-maintained container and is not an official Guide Labs image. It preserves the public upstream example while pinning the model, source, and runtime dependencies needed to reproduce it.

The image does not contain or redistribute model weights. It downloads the creator-owned public weights at runtime, prints the generated text and run metadata as JSON, and writes the same result to /mnt/output/steerling.json.

Change the prompt

Pass generation arguments after --:

anycloud job ghcr.io/anycloud-sh/steerling-8b:0.2.0-f2e5952-r2 \
  --gpu-type a10 --gpus all -- \
  --prompt "The future of interpretable models is" \
  --max-new-tokens 64 --steps 64 --seed 42

Use an AnyCloud output bucket to retain steerling.json after the job exits.

Reproduce the image

docker buildx build \
  --platform linux/amd64 \
  --build-arg IMAGE_REVISION="$(git rev-parse HEAD)" \
  --provenance=false \
  -t ghcr.io/anycloud-sh/steerling-8b:0.2.0-f2e5952-r2 \
  .

The exact published digest was run successfully on a Lambda NVIDIA A10. The deployment, input, output, GPU, and pinned revisions are recorded in validation/lambda-a10.json.

Pinned upstream inputs

• Steerling source: f34ffa89e46969445f3cf6e7c885e9623a2047c1
• Steerling-8B model: f2e59522c2f6d91c063770a4ce5514947a7b68bc
• Python package: steerling==0.2.0
• Minimum GPU memory enforced by the wrapper: 18 GiB

The upstream source is Apache-2.0 licensed. Review the model card for intended use and upstream-data considerations.