eval_results/AI-MO/mistral-7b-sft/aimo_v01.14/gsm8k/results_2024-04-24T13-32-58.138735.json

{
  "config_general": {
    "lighteval_sha": "?",
    "num_fewshot_seeds": 1,
    "override_batch_size": 4,
    "max_samples": null,
    "job_id": "",
    "start_time": 2347080.454418109,
    "end_time": 2347495.996548182,
    "total_evaluation_time_secondes": "415.5421300730668",
    "model_name": "AI-MO/mistral-7b-sft",
    "model_sha": "29e6b5eabf8b53c7fc8ef31774769de07a2fac05",
    "model_dtype": "torch.bfloat16",
    "model_size": "12.93 GB",
    "config": null
  },
  "results": {
    "lighteval|gsm8k|5": {
      "qem": 0.6413949962092494,
      "qem_stderr": 0.013210317364134035
    },
    "all": {
      "qem": 0.6413949962092494,
      "qem_stderr": 0.013210317364134035
    }
  },
  "versions": {
    "lighteval|gsm8k|5": 0
  },
  "config_tasks": {
    "lighteval|gsm8k": {
      "name": "gsm8k",
      "prompt_function": "gsm8k",
      "hf_repo": "gsm8k",
      "hf_subset": "main",
      "metric": [
        "quasi_exact_match_gsm8k"
      ],
      "hf_avail_splits": [
        "train",
        "test"
      ],
      "evaluation_splits": [
        "test"
      ],
      "few_shots_split": null,
      "few_shots_select": "random_sampling_from_train",
      "generation_size": 256,
      "stop_sequence": [
        "Question:"
      ],
      "output_regex": null,
      "frozen": false,
      "suite": [
        "lighteval"
      ],
      "original_num_docs": 1319,
      "effective_num_docs": 1319,
      "trust_dataset": true,
      "must_remove_duplicate_docs": null
    }
  },
  "summary_tasks": {
    "lighteval|gsm8k|5": {
      "hashes": {
        "hash_examples": "0ed016e24e7512fd",
        "hash_full_prompts": "b6d97958f0ed1df8",
        "hash_input_tokens": "3f58ee5e568409ec",
        "hash_cont_tokens": "4e8f0a3719b75fc1"
      },
      "truncated": 1319,
      "non_truncated": 0,
      "padded": 611,
      "non_padded": 708,
      "effective_few_shots": 5.0,
      "num_truncated_few_shots": 0
    }
  },
  "summary_general": {
    "hashes": {
      "hash_examples": "bc71463e88551d0e",
      "hash_full_prompts": "b5a5dacaed0816ff",
      "hash_input_tokens": "e4e0fd8eba82bd47",
      "hash_cont_tokens": "ec849dfb47a1032f"
    },
    "truncated": 1319,
    "non_truncated": 0,
    "padded": 611,
    "non_padded": 708,
    "num_truncated_few_shots": 0
  }
}