Setup & the O(n²) wall·2 min
Run the grader
0 / 1
# grade this stage (from your course directory)
npx theinference test --stage run-the-grader --watchBefore you write any cache code, run the harness once to prove the toolchain works and the GPT-2 weights are present. This is the loop you'll use on every build step: edit, run, read the diff.
Run it
npx theinference login # once — approves in your browser
npx theinference test --stage run-the-grader --watch
You should see setup ok. That confirms Node can run the starter, the GPT-2 weights load, and the grader can reach the fixtures.
No new download, no API key
Reuses the Track-1 GPT-2 124M weights (~250 MB safetensors) already in the image — no new download. CPU-only, no API key. The starter ships a known-good non-cached GPT-2 (your baseline to refactor) in src/gpt2.ts, in case you lost your llm-inference work — so you can start from a green forward pass either way.