eval_results/deepseek-ai/DeepSeek-R1-Distill-Qwen-32B/main/gpqa/results_2025-02-07T20-59-47.787665.json

{
  "config_general": {
    "lighteval_sha": "?",
    "num_fewshot_seeds": 1,
    "override_batch_size": -1,
    "max_samples": null,
    "job_id": 0,
    "start_time": 254887.031913544,
    "end_time": 255994.904675475,
    "total_evaluation_time_secondes": "1107.8727619310084",
    "model_name": "deepseek-ai/DeepSeek-R1-Distill-Qwen-32B",
    "model_sha": "",
    "model_dtype": null,
    "model_size": null
  },
  "results": {
    "custom|gpqa:diamond|0": {
      "extractive_match": 0.6313131313131313,
      "extractive_match_stderr": 0.03437305501980619
    },
    "all": {
      "extractive_match": 0.6313131313131313,
      "extractive_match_stderr": 0.03437305501980619
    }
  },
  "versions": {
    "custom|gpqa:diamond|0": 1
  },
  "config_tasks": {
    "custom|gpqa:diamond": {
      "name": "gpqa:diamond",
      "prompt_function": "gpqa_prompt_fn",
      "hf_repo": "Idavidrein/gpqa",
      "hf_subset": "gpqa_diamond",
      "metric": [
        {
          "metric_name": "extractive_match",
          "higher_is_better": true,
          "category": "3",
          "use_case": "1",
          "sample_level_fn": "sample_level_fn",
          "corpus_level_fn": "mean"
        }
      ],
      "hf_revision": null,
      "hf_filter": null,
      "hf_avail_splits": [
        "train"
      ],
      "trust_dataset": true,
      "evaluation_splits": [
        "train"
      ],
      "few_shots_split": null,
      "few_shots_select": null,
      "generation_size": 32768,
      "generation_grammar": null,
      "stop_sequence": [],
      "num_samples": null,
      "suite": [
        "custom"
      ],
      "original_num_docs": 198,
      "effective_num_docs": 198,
      "must_remove_duplicate_docs": false,
      "version": 1
    }
  },
  "summary_tasks": {
    "custom|gpqa:diamond|0": {
      "hashes": {
        "hash_examples": "50ecb6f5d091bd95",
        "hash_full_prompts": "4d6bc2c8e64a03b8",
        "hash_input_tokens": "7f4457760a5d7f38",
        "hash_cont_tokens": "833aac14af8800eb"
      },
      "truncated": 0,
      "non_truncated": 198,
      "padded": 0,
      "non_padded": 198,
      "effective_few_shots": 0.0,
      "num_truncated_few_shots": 0
    }
  },
  "summary_general": {
    "hashes": {
      "hash_examples": "a9318dbdd867770b",
      "hash_full_prompts": "d8f2b1ad973f6d42",
      "hash_input_tokens": "1edd6765f01885b1",
      "hash_cont_tokens": "0f0e646baaaeb748"
    },
    "truncated": 0,
    "non_truncated": 198,
    "padded": 0,
    "non_padded": 198,
    "num_truncated_few_shots": 0
  }
}