Skip to content

Run a synthetic smoke test

Kind: how-to.

Use this procedure from the repository root. Follow the GPU prerequisite instructions first. The serving profile needs at least 11000 MiB of free GPU memory.

  1. Install the client and development tools.
uv sync --locked
  1. Generate the eight synthetic page images.
uv run inference-axi corpus
  1. Build the server image.
uv run inference-axi build
  1. Check the available GPU memory.
uv run inference-axi status
  1. Start the demo server.
uv run inference-axi start
  1. Check one page with known text.
uv run inference-axi ocr corpus/page-01.png --output runs/page-01.txt --expect LAB401
uv run pytest -m gpu
  1. Run the four concurrency levels.
uv run inference-axi bench corpus --output runs/first

  1. Read runs/first/summary.md.

  2. Check the failures and raw outputs in runs/first/results.json.

  3. Remove the demo container.

    uv run inference-axi stop

Use uv run inference-axi logs if startup fails. Use uv run inference-axi logs --full for all log lines. Run stop after a startup failure to remove the failed container.