eval_results/HuggingFaceH4/qwen-1.5-1.8b-dpo/v1.1/gsm8k/results_2024-03-26T12-06-36.364394.json

{
  "config_general": {
    "lighteval_sha": "?",
    "num_fewshot_seeds": 1,
    "override_batch_size": 1,
    "max_samples": null,
    "job_id": "",
    "start_time": 1887083.477384986,
    "end_time": 1887816.438484011,
    "total_evaluation_time_secondes": "732.9610990250949",
    "model_name": "HuggingFaceH4/qwen-1.5-1.8b-dpo",
    "model_sha": "16a3182fb99083685eda8253f019f4c93889e0d9",
    "model_dtype": "torch.bfloat16",
    "model_size": "3.79 GB",
    "config": null
  },
  "results": {
    "lighteval|gsm8k|5": {
      "qem": 0.2714177407126611,
      "qem_stderr": 0.01224900202615058
    }
  },
  "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
    }
  },
  "summary_tasks": {
    "lighteval|gsm8k|5": {
      "hashes": {
        "hash_examples": "0ed016e24e7512fd",
        "hash_full_prompts": "a9c6515c2ecd2d07",
        "hash_input_tokens": "cf5e91d1c1d6ca10",
        "hash_cont_tokens": "f48d1a67cfa0c7f1"
      },
      "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": "bc71463e88551d0e",
      "hash_full_prompts": "f49691ec139bfeac",
      "hash_input_tokens": "4db9bb980ae67af7",
      "hash_cont_tokens": "0caf47b3ed8965e5"
    },
    "truncated": 0,
    "non_truncated": 1319,
    "padded": 0,
    "non_padded": 1319,
    "num_truncated_few_shots": 0
  }
}