aggregate_results.json

{
  "model": "Qwen/Qwen2.5-0.5B-Instruct",
  "n_requested_per_task": 8,
  "temperatures": [
    1.0,
    2.0
  ],
  "methods": [
    "top_p",
    "min_p",
    "top_h",
    "top_w"
  ],
  "runtime_seconds": 159.78445148468018,
  "aggregates": [
    {
      "task": "gsm8k",
      "temperature": 1.0,
      "method": "top_p",
      "correct": 1,
      "n": 8,
      "accuracy": 0.125,
      "wilson_low": 0.02241690886329617,
      "wilson_high": 0.4708948057698615
    },
    {
      "task": "gsm8k",
      "temperature": 1.0,
      "method": "min_p",
      "correct": 0,
      "n": 8,
      "accuracy": 0.0,
      "wilson_low": 0.0,
      "wilson_high": 0.3244156195108769
    },
    {
      "task": "gsm8k",
      "temperature": 1.0,
      "method": "top_h",
      "correct": 1,
      "n": 8,
      "accuracy": 0.125,
      "wilson_low": 0.02241690886329617,
      "wilson_high": 0.4708948057698615
    },
    {
      "task": "gsm8k",
      "temperature": 1.0,
      "method": "top_w",
      "correct": 0,
      "n": 8,
      "accuracy": 0.0,
      "wilson_low": 0.0,
      "wilson_high": 0.3244156195108769
    },
    {
      "task": "gsm8k",
      "temperature": 2.0,
      "method": "top_p",
      "correct": 1,
      "n": 8,
      "accuracy": 0.125,
      "wilson_low": 0.02241690886329617,
      "wilson_high": 0.4708948057698615
    },
    {
      "task": "gsm8k",
      "temperature": 2.0,
      "method": "min_p",
      "correct": 0,
      "n": 8,
      "accuracy": 0.0,
      "wilson_low": 0.0,
      "wilson_high": 0.3244156195108769
    },
    {
      "task": "gsm8k",
      "temperature": 2.0,
      "method": "top_h",
      "correct": 0,
      "n": 8,
      "accuracy": 0.0,
      "wilson_low": 0.0,
      "wilson_high": 0.3244156195108769
    },
    {
      "task": "gsm8k",
      "temperature": 2.0,
      "method": "top_w",
      "correct": 1,
      "n": 8,
      "accuracy": 0.125,
      "wilson_low": 0.02241690886329617,
      "wilson_high": 0.4708948057698615
    }
  ]
}