{
 "file": "gpu-ql3090-24g-nemo-q3-k-m-control-3090-q3_K_M-ladder.json",
 "model_tag": "mistral-nemo:12b-instruct-2407-q3_K_M",
 "arm": "control-3090",
 "box_class": "gpu-ql3090-24g",
 "quantization": "Q3_K_M",
 "store_gib": 5.67,
 "largest_whole": 131072,
 "topped_out": true,
 "stopped_at": null,
 "errors": [],
 "ups_before": 13.0,
 "ups_after": 20.0,
 "windows": [
  {
   "num_ctx": 4096,
   "whole": true,
   "verdict": "fits: fully resident on the card",
   "vram_delta_mib": {
    "<withheld: GPU UUID of board A>": 6291.0
   },
   "runs": 3,
   "decode_median": 74.894,
   "ttft_ms_median": 510.56,
   "board_w_median": 257.59,
   "j_per_1k_median": 3439.41,
   "sw_power_cap_fraction_median": 1.0,
   "temp_max_c": 60.0,
   "spread": 0.0053,
   "decode_range": 0.398,
   "throttle": [],
   "stop": null,
   "gate_refusals": 0,
   "gates_failed": [],
   "passes": true,
   "headline_decode": 74.894
  },
  {
   "num_ctx": 8192,
   "whole": true,
   "verdict": "fits: fully resident on the card",
   "vram_delta_mib": {
    "<withheld: GPU UUID of board A>": 470.0
   },
   "runs": 3,
   "decode_median": 74.626,
   "ttft_ms_median": 515.71,
   "board_w_median": 262.99,
   "j_per_1k_median": 3524.1,
   "sw_power_cap_fraction_median": 1.0,
   "temp_max_c": 64.0,
   "spread": 0.006,
   "decode_range": 0.451,
   "throttle": [],
   "stop": null,
   "gate_refusals": 0,
   "gates_failed": [],
   "passes": true,
   "headline_decode": 74.626
  },
  {
   "num_ctx": 16384,
   "whole": true,
   "verdict": "fits: fully resident on the card",
   "vram_delta_mib": {
    "<withheld: GPU UUID of board A>": 696.0
   },
   "runs": 3,
   "decode_median": 74.415,
   "ttft_ms_median": 519.92,
   "board_w_median": 266.35,
   "j_per_1k_median": 3579.26,
   "sw_power_cap_fraction_median": 1.0,
   "temp_max_c": 64.0,
   "spread": 0.0036,
   "decode_range": 0.269,
   "throttle": [],
   "stop": null,
   "gate_refusals": 0,
   "gates_failed": [],
   "passes": true,
   "headline_decode": 74.415
  },
  {
   "num_ctx": 32768,
   "whole": true,
   "verdict": "fits: fully resident on the card",
   "vram_delta_mib": {
    "<withheld: GPU UUID of board A>": 1392.0
   },
   "runs": 3,
   "decode_median": 74.135,
   "ttft_ms_median": 538.62,
   "board_w_median": 262.38,
   "j_per_1k_median": 3545.24,
   "sw_power_cap_fraction_median": 1.0,
   "temp_max_c": 65.0,
   "spread": 0.0036,
   "decode_range": 0.266,
   "throttle": [],
   "stop": null,
   "gate_refusals": 0,
   "gates_failed": [],
   "passes": true,
   "headline_decode": 74.135
  },
  {
   "num_ctx": 49152,
   "whole": true,
   "verdict": "fits: fully resident on the card",
   "vram_delta_mib": {
    "<withheld: GPU UUID of board A>": 1432.0
   },
   "runs": 3,
   "decode_median": 74.037,
   "ttft_ms_median": 535.74,
   "board_w_median": 261.63,
   "j_per_1k_median": 3534.41,
   "sw_power_cap_fraction_median": 1.0,
   "temp_max_c": 64.0,
   "spread": 0.0003,
   "decode_range": 0.02,
   "throttle": [],
   "stop": null,
   "gate_refusals": 0,
   "gates_failed": [],
   "passes": true,
   "headline_decode": 74.037
  },
  {
   "num_ctx": 65536,
   "whole": true,
   "verdict": "fits: fully resident on the card",
   "vram_delta_mib": {
    "<withheld: GPU UUID of board A>": 1350.0
   },
   "runs": 3,
   "decode_median": 74.245,
   "ttft_ms_median": 520.47,
   "board_w_median": 265.6,
   "j_per_1k_median": 3577.33,
   "sw_power_cap_fraction_median": 1.0,
   "temp_max_c": 63.0,
   "spread": 0.0005,
   "decode_range": 0.035,
   "throttle": [],
   "stop": null,
   "gate_refusals": 0,
   "gates_failed": [],
   "passes": true,
   "headline_decode": 74.245
  },
  {
   "num_ctx": 98304,
   "whole": true,
   "verdict": "fits: fully resident on the card",
   "vram_delta_mib": {
    "<withheld: GPU UUID of board A>": 2880.0
   },
   "runs": 3,
   "decode_median": 74.087,
   "ttft_ms_median": 480.5,
   "board_w_median": 267.91,
   "j_per_1k_median": 3620.29,
   "sw_power_cap_fraction_median": 1.0,
   "temp_max_c": 64.0,
   "spread": 0.0018,
   "decode_range": 0.137,
   "throttle": [],
   "stop": null,
   "gate_refusals": 0,
   "gates_failed": [],
   "passes": true,
   "headline_decode": 74.087
  },
  {
   "num_ctx": 131072,
   "whole": true,
   "verdict": "fits: fully resident on the card",
   "vram_delta_mib": {
    "<withheld: GPU UUID of board A>": 2880.0
   },
   "runs": 3,
   "decode_median": 74.042,
   "ttft_ms_median": 525.66,
   "board_w_median": 266.47,
   "j_per_1k_median": 3598.89,
   "sw_power_cap_fraction_median": 1.0,
   "temp_max_c": 64.0,
   "spread": 0.0007,
   "decode_range": 0.051,
   "throttle": [],
   "stop": null,
   "gate_refusals": 0,
   "gates_failed": [],
   "passes": true,
   "headline_decode": 74.042
  }
 ],
 "resident_named_card_growth_mib": 6291.0,
 "resident": true,
 "passing_windows": [
  4096,
  8192,
  16384,
  32768,
  49152,
  65536,
  98304,
  131072
 ]
}