{
 "load_before": 0.43,
 "reps": 5,
 "note": "round-robin interleaved; median of REPS; llama-bench -r 1 per invocation",
 "A_7b_thread_sweep": [
  {
   "id": "7B t=1",
   "threads": 1,
   "pp": 128,
   "tg": 32,
   "prefill_median": 28.89,
   "prefill_stdev": 0.14,
   "prefill_n": 5,
   "decode_median": 7.26,
   "decode_stdev": 0.2,
   "decode_n": 5
  },
  {
   "id": "7B t=2",
   "threads": 2,
   "pp": 128,
   "tg": 32,
   "prefill_median": 55.41,
   "prefill_stdev": 0.78,
   "prefill_n": 5,
   "decode_median": 12.95,
   "decode_stdev": 0.13,
   "decode_n": 5
  },
  {
   "id": "7B t=4",
   "threads": 4,
   "pp": 128,
   "tg": 32,
   "prefill_median": 104.42,
   "prefill_stdev": 0.92,
   "prefill_n": 5,
   "decode_median": 17.76,
   "decode_stdev": 2.04,
   "decode_n": 5
  },
  {
   "id": "7B t=8",
   "threads": 8,
   "pp": 128,
   "tg": 32,
   "prefill_median": 179.45,
   "prefill_stdev": 3.76,
   "prefill_n": 5,
   "decode_median": 24.45,
   "decode_stdev": 1.05,
   "decode_n": 5
  },
  {
   "id": "7B t=16",
   "threads": 16,
   "pp": 128,
   "tg": 32,
   "prefill_median": 212.41,
   "prefill_stdev": 1.27,
   "prefill_n": 5,
   "decode_median": 21.01,
   "decode_stdev": 0.75,
   "decode_n": 5
  },
  {
   "id": "7B t=20",
   "threads": 20,
   "pp": 128,
   "tg": 32,
   "prefill_median": 218.39,
   "prefill_stdev": 2.49,
   "prefill_n": 5,
   "decode_median": 18.03,
   "decode_stdev": 1.05,
   "decode_n": 5
  }
 ],
 "A_05b_thread_sweep": [
  {
   "id": "0.5B t=1",
   "threads": 1,
   "pp": 128,
   "tg": 32,
   "prefill_median": 318.95,
   "prefill_stdev": 6.43,
   "prefill_n": 5,
   "decode_median": 85.41,
   "decode_stdev": 2.93,
   "decode_n": 5
  },
  {
   "id": "0.5B t=2",
   "threads": 2,
   "pp": 128,
   "tg": 32,
   "prefill_median": 556.71,
   "prefill_stdev": 5.35,
   "prefill_n": 5,
   "decode_median": 138.6,
   "decode_stdev": 9.15,
   "decode_n": 5
  },
  {
   "id": "0.5B t=4",
   "threads": 4,
   "pp": 128,
   "tg": 32,
   "prefill_median": 915.65,
   "prefill_stdev": 11.77,
   "prefill_n": 5,
   "decode_median": 177.61,
   "decode_stdev": 13.21,
   "decode_n": 5
  },
  {
   "id": "0.5B t=8",
   "threads": 8,
   "pp": 128,
   "tg": 32,
   "prefill_median": 1457.68,
   "prefill_stdev": 21.84,
   "prefill_n": 5,
   "decode_median": 190.32,
   "decode_stdev": 8.93,
   "decode_n": 5
  },
  {
   "id": "0.5B t=16",
   "threads": 16,
   "pp": 128,
   "tg": 32,
   "prefill_median": 1047.66,
   "prefill_stdev": 51.97,
   "prefill_n": 5,
   "decode_median": 120.05,
   "decode_stdev": 7.16,
   "decode_n": 5
  },
  {
   "id": "0.5B t=20",
   "threads": 20,
   "pp": 128,
   "tg": 32,
   "prefill_median": 965.98,
   "prefill_stdev": 87.36,
   "prefill_n": 5,
   "decode_median": 45.49,
   "decode_stdev": 6.43,
   "decode_n": 5
  }
 ],
 "B_finding3_cost": [
  {
   "id": "0.5B BROKEN build",
   "pp": 128,
   "tg": 32,
   "prefill_median": 657.0,
   "prefill_stdev": 51.29,
   "prefill_n": 5,
   "decode_median": 42.18,
   "decode_stdev": 6.78,
   "decode_n": 5
  },
  {
   "id": "0.5B FIXED build",
   "pp": 128,
   "tg": 32,
   "prefill_median": 933.63,
   "prefill_stdev": 74.03,
   "prefill_n": 5,
   "decode_median": 41.7,
   "decode_stdev": 6.86,
   "decode_n": 5
  },
  {
   "id": "7B BROKEN build",
   "pp": 128,
   "tg": 32,
   "prefill_median": 48.64,
   "prefill_stdev": 0.42,
   "prefill_n": 5,
   "decode_median": 11.17,
   "decode_stdev": 0.37,
   "decode_n": 5
  },
  {
   "id": "7B FIXED build",
   "pp": 128,
   "tg": 32,
   "prefill_median": 222.14,
   "prefill_stdev": 4.94,
   "prefill_n": 5,
   "decode_median": 18.45,
   "decode_stdev": 0.84,
   "decode_n": 5
  }
 ],
 "load_after": 12.32
}