eval-results/PulsarAI/EnsembleV5-Nova-13B/results_2023-10-23T15-24-00.966689.json

{
  "config_general": {
    "model_name": "PulsarAI/EnsembleV5-Nova-13B",
    "model_sha": "b2b03e4a3714d5c738bb49767c1945da4f4d98d6",
    "model_size": "24.32 GB",
    "model_dtype": "torch.float16",
    "lighteval_sha": "0f318ecf002208468154899217b3ba7c6ae09374",
    "num_few_shot_default": 0,
    "num_fewshot_seeds": 1,
    "override_batch_size": 1,
    "max_samples": null,
    "job_id": ""
  },
  "results": {
    "harness|drop|3": {
      "em": 0.007445469798657718,
      "em_stderr": 0.0008803652515899855,
      "f1": 0.08636220637583875,
      "f1_stderr": 0.0018310737230495444
    },
    "harness|gsm8k|5": {
      "acc": 0.10765731614859743,
      "acc_stderr": 0.008537484003023352
    },
    "harness|winogrande|5": {
      "acc": 0.7624309392265194,
      "acc_stderr": 0.011961298905803157
    },
    "all": {
      "em": 0.007445469798657718,
      "em_stderr": 0.0008803652515899855,
      "f1": 0.08636220637583875,
      "f1_stderr": 0.0018310737230495444,
      "acc": 0.4350441276875584,
      "acc_stderr": 0.010249391454413254
    }
  },
  "versions": {
    "harness|drop|3": 1,
    "harness|gsm8k|5": 0,
    "harness|winogrande|5": 0,
    "all": 0
  },
  "config_tasks": {
    "harness|drop": "LM Harness task",
    "harness|gsm8k": "LM Harness task",
    "harness|winogrande": "LM Harness task"
  },
  "summary_tasks": {
    "harness|drop|3": {
      "hashes": {
        "hash_examples": "1d27416e8324e9a3",
        "hash_full_prompts": "a5513ff9a741b385",
        "hash_input_tokens": "42076f0efbb50aa6",
        "hash_cont_tokens": "3cf8d80a0a153ca6"
      },
      "truncated": 3,
      "non-truncated": 9533,
      "padded": 0,
      "non-padded": 9536,
      "effective_few_shots": 3.0,
      "num_truncated_few_shots": 0
    },
    "harness|gsm8k|5": {
      "hashes": {
        "hash_examples": "4c0843a5d99bcfdc",
        "hash_full_prompts": "41d55e83abc0e02d",
        "hash_input_tokens": "bda342e47b5099b2",
        "hash_cont_tokens": "e4063189c38e1b54"
      },
      "truncated": 0,
      "non-truncated": 1319,
      "padded": 0,
      "non-padded": 1319,
      "effective_few_shots": 5.0,
      "num_truncated_few_shots": 0
    },
    "harness|winogrande|5": {
      "hashes": {
        "hash_examples": "aada0a176fd81218",
        "hash_full_prompts": "c8655cbd12de8409",
        "hash_input_tokens": "c0bedf98cb040854",
        "hash_cont_tokens": "f08975ad6f2d5864"
      },
      "truncated": 0,
      "non-truncated": 2534,
      "padded": 2432,
      "non-padded": 102,
      "effective_few_shots": 5.0,
      "num_truncated_few_shots": 0
    }
  },
  "summary_general": {
    "hashes": {
      "hash_examples": "9b4d8993161e637d",
      "hash_full_prompts": "08215e527b7e60a5",
      "hash_input_tokens": "a12f3e3c934bd78b",
      "hash_cont_tokens": "0852126b435f5e70"
    },
    "total_evaluation_time_secondes": "42982.43106532097",
    "truncated": 3,
    "non-truncated": 13386,
    "padded": 2432,
    "non-padded": 10957,
    "num_truncated_few_shots": 0
  }
}