cookbook-add-model

1// TEMPLATE — instantiate via the cookbook-add-model skill. NOT a live cookbook.2// Copy to docs/src/snippets/configs/<hf-org>/<model-slug>-benchmarks.jsx and3// fill measured numbers — OR delete this file entirely if you have none yet (the4// MDX simply omits the `benchmarks` import/prop).5//6// One entry per cell `match` tuple (same 5 keys as config cells). The card stays7// "pending" until an entry has a non-null speed metric or accuracy. Speed shape:8//   speed: [{ workload: {dataset, isl, osl, max_concurrency}, ttft_ms, tpot_ms,9//             tokens_per_sec_per_gpu }, ...]10//   - ttft_ms/tpot_ms are P50 (median); set config.latencyPercentile ("P50" default, or "Mean");11//     an entry-level latencyPercentile overrides the page value per cell.12//   - tokens_per_sec_per_gpu = total (in+out) tok/s/GPU13//     (= output tok/s ÷ GPUs × (isl+osl)/osl). interactivity is derived = 1000/TPOT (tokens/s/user).14// Per-cell `accuracy: { <key>: <pct> }` overrides the config's defaultAccuracy.15 16export const benchmarks = [17  // EXAMPLE — one filled entry showing the shape; replace numbers, add one per cell.18  {19    match: { hw: "b200", variant: "default", quant: "fp4", strategy: "low-latency", nodes: "single" },20    sglang_version: "0.0.0", // TODO: ASK the user for the sglang version these numbers were measured on — don't invent one21    speed: [22      { workload: { dataset: "random", isl: 8192, osl: 1024, max_concurrency: 1 },23        ttft_ms: null, tpot_ms: null, tokens_per_sec_per_gpu: null },24      { workload: { dataset: "random", isl: 8192, osl: 1024, max_concurrency: 16 },25        ttft_ms: null, tpot_ms: null, tokens_per_sec_per_gpu: null },26    ],27  },28  // Bare-match stubs (no data yet) are fine — the card shows "pending" for these.29  { match: { hw: "h200",   variant: "default", quant: "fp8",  strategy: "balanced",        nodes: "single"  } },30  { match: { hw: "h100",   variant: "default", quant: "fp4",  strategy: "high-throughput", nodes: "single"  } },31  { match: { hw: "mi300x", variant: "default", quant: "bf16", strategy: "balanced",        nodes: "single"  } },32  { match: { hw: "b200",   variant: "default", quant: "fp4",  strategy: "high-throughput", nodes: "multi-2" } },33];34