eval-results/AlekseyKorshuk/pygmalion-6b-vicuna-chatml/results_2023-12-02T15-52-11.848314.json

{
  "config_general": {
    "lighteval_sha": "b35d4d84573be82d91c07ea46260f262f72cf69d",
    "num_few_shot_default": 0,
    "num_fewshot_seeds": 1,
    "override_batch_size": 1,
    "max_samples": null,
    "job_id": "",
    "start_time": 1390201.169224904,
    "end_time": 1392498.943953603,
    "total_evaluation_time_secondes": "2297.7747286991216",
    "model_name": "AlekseyKorshuk/pygmalion-6b-vicuna-chatml",
    "model_sha": "ee3ada91a69a194cedfabbfeab98f1499b75cb44",
    "model_dtype": "torch.float16",
    "model_size": "11.28 GB"
  },
  "results": {
    "harness|gsm8k|5": {
      "acc": 0.04397270659590599,
      "acc_stderr": 0.005647666449126458
    },
    "all": {
      "acc": 0.04397270659590599,
      "acc_stderr": 0.005647666449126458
    }
  },
  "versions": {
    "all": 0,
    "harness|gsm8k|5": 0
  },
  "config_tasks": {
    "harness|gsm8k": "LM Harness task"
  },
  "summary_tasks": {
    "harness|gsm8k|5": {
      "hashes": {
        "hash_examples": "4c0843a5d99bcfdc",
        "hash_full_prompts": "41d55e83abc0e02d",
        "hash_input_tokens": "3ab9b4c5105492a3",
        "hash_cont_tokens": "f191a21e6082b576"
      },
      "truncated": 0,
      "non_truncated": 1319,
      "padded": 0,
      "non_padded": 1319,
      "effective_few_shots": 5.0,
      "num_truncated_few_shots": 0
    }
  },
  "summary_general": {
    "hashes": {
      "hash_examples": "18b756b7813d1bdf",
      "hash_full_prompts": "deb3b1dff10b95aa",
      "hash_input_tokens": "ce30a215cb9fb6d1",
      "hash_cont_tokens": "1685e98f0fcdfe91"
    },
    "truncated": 0,
    "non_truncated": 1319,
    "padded": 0,
    "non_padded": 1319,
    "num_truncated_few_shots": 0
  }
}