{
  "bench": "power-cap-96gb",
  "arm": "ollama-q4-mistral",
  "mode": "go",
  "started_utc": "2026-08-25T10:48:41Z",
  "ended_utc": "2026-08-25T10:52:43Z",
  "endpoint": "<serving host>",
  "engine": "ollama",
  "model_requested": "mistral-small3.2:24b",
  "models_served": [
    "gemma4:26b-a4b-it-bf16",
    "mistral-small3.2:24b-instruct-2506-q8_0",
    "gemma4:26b-a4b-it-q8_0",
    "llama3.3:70b",
    "nemotron3:33b",
    "granite4.1:30b-q8_0",
    "nomic-embed-text:latest",
    "mistral-small3.2:24b",
    "gemma4:26b"
  ],
  "prompt_file": "p512.txt",
  "prompt_sha256": "90eedd0c53f9554ae3837674504fcb7090013d9a432a653352183c0f25a7ce5c",
  "prompt_tokens_measured": null,
  "max_tokens": 256,
  "ignore_eos": false,
  "thinking": "off",
  "temperature": 0.0,
  "seed": 20260825,
  "repeats": 3,
  "warmup_waves": 1,
  "power_window": {
    "samples": 483,
    "power_w_mean": 594.6,
    "power_w_peak": 603.5,
    "vram_mib_peak": 38794.0,
    "csv": "power.csv"
  },
  "rows": [
    {
      "concurrency": 1,
      "waves": 3,
      "requests_ok": 3,
      "requests_failed": 0,
      "ttft_ms_p50": 25.9,
      "ttft_ms_p95": 26.0,
      "ttft_ms_min": 24.2,
      "ttft_ms_max": 26.0,
      "decode_tok_s_per_stream_p50": 94.65,
      "decode_tok_s_per_stream_mean": 94.65,
      "aggregate_tok_s_mean": 94.1,
      "aggregate_tok_s_max": 94.14,
      "completion_tokens_seen": 768,
      "reasoning_chunks_seen": 0,
      "answer_chars_seen": 3114,
      "distinct_reply_shas": 1
    },
    {
      "concurrency": 4,
      "waves": 3,
      "requests_ok": 12,
      "requests_failed": 0,
      "ttft_ms_p50": 5446.8,
      "ttft_ms_p95": 8170.2,
      "ttft_ms_min": 20.9,
      "ttft_ms_max": 8187.0,
      "decode_tok_s_per_stream_p50": 94.52,
      "decode_tok_s_per_stream_mean": 94.41,
      "aggregate_tok_s_mean": 94.14,
      "aggregate_tok_s_max": 94.26,
      "completion_tokens_seen": 3072,
      "reasoning_chunks_seen": 0,
      "answer_chars_seen": 12456,
      "distinct_reply_shas": 1
    },
    {
      "concurrency": 8,
      "waves": 3,
      "requests_ok": 24,
      "requests_failed": 0,
      "ttft_ms_p50": 10908.4,
      "ttft_ms_p95": 19100.1,
      "ttft_ms_min": 24.3,
      "ttft_ms_max": 19184.3,
      "decode_tok_s_per_stream_p50": 94.3,
      "decode_tok_s_per_stream_mean": 94.09,
      "aggregate_tok_s_mean": 93.84,
      "aggregate_tok_s_max": 94.11,
      "completion_tokens_seen": 6144,
      "reasoning_chunks_seen": 0,
      "answer_chars_seen": 24912,
      "distinct_reply_shas": 1
    },
    {
      "concurrency": 16,
      "waves": 3,
      "requests_ok": 48,
      "requests_failed": 0,
      "ttft_ms_p50": 21906.9,
      "ttft_ms_p95": 41046.3,
      "ttft_ms_min": 26.6,
      "ttft_ms_max": 41058.3,
      "decode_tok_s_per_stream_p50": 93.79,
      "decode_tok_s_per_stream_mean": 93.82,
      "aggregate_tok_s_mean": 93.58,
      "aggregate_tok_s_max": 93.62,
      "completion_tokens_seen": 12288,
      "reasoning_chunks_seen": 0,
      "answer_chars_seen": 49824,
      "distinct_reply_shas": 1
    }
  ],
  "provenance": {
    "harness_sha256": "96a8f38dd1bbb89ce24f89339d81019fd91435e06e7be21b8476f76ba9ce857b",
    "harness_host": "<harness host>",
    "python": "3.14.4",
    "platform": "Linux-7.0.0-30-generic-x86_64-with-glibc2.43"
  },
  "_kit_note": "power.csv and throttle.csv in this directory carry UTC, Z-stamped timestamps; see the kit README for what was converted and why."
}