LLM 评测流程. Copy the source, paste into the official playground, done. No coordinates — every position is stated relative to something else.
node cases "Test cases"
node run "Eval runner" below cases
node model "Candidate model" right of run
node base "Baseline model" right of model
node judge "LLM judge" below run
node report "Score report" below judge
edge cases -> run
edge model -> judge
edge base -> judge
edge judge -> reportnpx reladraw llm-eval-flow.reladraw -o out.svgTest cases through a runner, candidate and baseline both scored by an LLM judge into one report.. Use it when teams argue about model choice — the judge node makes comparisons reproducible.
node new_id 'New label' below anchor — placement is relative, so the solver re-flows the whole diagram instantly.edge a -> b 'your label' — labels are the fastest way to carry rates, protocols or percentages.below to right of rotates a branch sideways without touching any other line.Declare each box with node, connect them with edge, and state one position per node relative to another. Paste the source from this page into the official playground and it renders immediately.
Yes — the source is free to copy, modify and ship in docs, READMEs, books or client decks. No attribution required.
Render in the playground and export SVG, or run it locally: npx reladraw file.reladraw -o out.svg for CI-friendly builds.