meta-llama/Llama-2-70b-chat-hf/results_2023-10-17T05-07-42.486452.json

{
  "config_general": {
    "model_name": "meta-llama/Llama-2-70b-chat-hf",
    "model_sha": "cfe96d938c52db7c6d936f99370c0801b24233c4",
    "model_size": "128.64 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": 0.040373322147651006,
      "em_stderr": 0.0020157564185176837,
      "f1": 0.1050272651006715,
      "f1_stderr": 0.0023756238577676155
    },
    "harness|gsm8k|5": {
      "acc": 0.266868840030326,
      "acc_stderr": 0.012183780551887957
    },
    "harness|winogrande|5": {
      "acc": 0.8050513022888713,
      "acc_stderr": 0.011134099415938268
    },
    "all": {
      "em": 0.040373322147651006,
      "em_stderr": 0.0020157564185176837,
      "f1": 0.1050272651006715,
      "f1_stderr": 0.0023756238577676155,
      "acc": 0.5359600711595986,
      "acc_stderr": 0.011658939983913113
    }
  },
  "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": "b7f7e4a7d842e431"
      },
      "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": "13bcb12a5f7991f1"
      },
      "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": "5fa49b6513c85264"
    },
    "total_evaluation_time_secondes": "42063.55081868172",
    "truncated": 3,
    "non-truncated": 13386,
    "padded": 2432,
    "non-padded": 10957,
    "num_truncated_few_shots": 0
  }
}