eval-results/TheBloke/Wizard-Vicuna-13B-Uncensored-GPTQ/results_2023-11-07T13-33-21.987098.json

{
  "config_general": {
    "lighteval_sha": "167773f1d5d1647c60dadc31c9e731ab7dbcbbad",
    "num_few_shot_default": 0,
    "num_fewshot_seeds": 1,
    "override_batch_size": 1,
    "max_samples": null,
    "job_id": "",
    "model_name": "TheBloke/Wizard-Vicuna-13B-Uncensored-GPTQ",
    "model_sha": "c322bec72dabdbf1fe4a8196b0234694b316a1a7",
    "model_dtype": "torch.float16",
    "model_size": "6.8 GB"
  },
  "results": {
    "harness|drop|3": {
      "em": 0.07581795302013423,
      "em_stderr": 0.0027108434788949637,
      "f1": 0.14551698825503265,
      "f1_stderr": 0.0029781067344765754
    },
    "harness|gsm8k|5": {
      "acc": 0.09931766489764973,
      "acc_stderr": 0.008238371412683973
    },
    "harness|winogrande|5": {
      "acc": 0.7576953433307024,
      "acc_stderr": 0.012042352526174787
    },
    "all": {
      "em": 0.07581795302013423,
      "em_stderr": 0.0027108434788949637,
      "f1": 0.14551698825503265,
      "f1_stderr": 0.0029781067344765754,
      "acc": 0.42850650411417607,
      "acc_stderr": 0.010140361969429381
    }
  },
  "versions": {
    "all": 0,
    "harness|drop|3": 1,
    "harness|gsm8k|5": 0,
    "harness|winogrande|5": 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": "61b608e0b5ceed76",
        "hash_cont_tokens": "0ee6629fc6a91da8"
      },
      "truncated": 1263,
      "non_truncated": 8273,
      "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": "dc55f4a29f149379"
      },
      "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": 1267,
      "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": "80afe720f936f8d2",
      "hash_cont_tokens": "e3ddc2074581a43b"
    },
    "truncated": 1263,
    "non_truncated": 10859,
    "padded": 2432,
    "non_padded": 10957,
    "num_truncated_few_shots": 0,
    "total_evaluation_time_secondes": 0
  }
}