4b284b12bc4/evaluation/generation/slim.4b284b12bc4_GEM-wiki_lingua_en_tldr_en_4.json

{
  "results": [
    {
      "task_name": "GEM/wiki_lingua_en",
      "prompt_name": "tldr_en",
      "rouge1_precision": 0.05323367871163294,
      "dataset_path": "GEM/wiki_lingua",
      "dataset_name": "en",
      "subset": null,
      "rouge1_precision_stderr": 0.002019764759552398
    },
    {
      "task_name": "GEM/wiki_lingua_en",
      "prompt_name": "tldr_en",
      "rouge1_recall": 0.08477832633294664,
      "dataset_path": "GEM/wiki_lingua",
      "dataset_name": "en",
      "subset": null,
      "rouge1_recall_stderr": 0.0029256767843087337
    },
    {
      "task_name": "GEM/wiki_lingua_en",
      "prompt_name": "tldr_en",
      "rouge1_fmeasure": 0.057194516095400834,
      "dataset_path": "GEM/wiki_lingua",
      "dataset_name": "en",
      "subset": null,
      "rouge1_fmeasure_stderr": 0.0019532780600384314
    },
    {
      "task_name": "GEM/wiki_lingua_en",
      "prompt_name": "tldr_en",
      "rouge2_precision": 0.011892613439611567,
      "dataset_path": "GEM/wiki_lingua",
      "dataset_name": "en",
      "subset": null,
      "rouge2_precision_stderr": 0.0006525291351364055
    },
    {
      "task_name": "GEM/wiki_lingua_en",
      "prompt_name": "tldr_en",
      "rouge2_recall": 0.021014952711012305,
      "dataset_path": "GEM/wiki_lingua",
      "dataset_name": "en",
      "subset": null,
      "rouge2_recall_stderr": 0.0011420293722952714
    },
    {
      "task_name": "GEM/wiki_lingua_en",
      "prompt_name": "tldr_en",
      "rouge2_fmeasure": 0.013407675922368708,
      "dataset_path": "GEM/wiki_lingua",
      "dataset_name": "en",
      "subset": null,
      "rouge2_fmeasure_stderr": 0.0006687476239191229
    },
    {
      "task_name": "GEM/wiki_lingua_en",
      "prompt_name": "tldr_en",
      "rougeL_precision": 0.03994947969161068,
      "dataset_path": "GEM/wiki_lingua",
      "dataset_name": "en",
      "subset": null,
      "rougeL_precision_stderr": 0.0015323136750776513
    },
    {
      "task_name": "GEM/wiki_lingua_en",
      "prompt_name": "tldr_en",
      "rougeL_recall": 0.0654114763710059,
      "dataset_path": "GEM/wiki_lingua",
      "dataset_name": "en",
      "subset": null,
      "rougeL_recall_stderr": 0.0022890377474177525
    },
    {
      "task_name": "GEM/wiki_lingua_en",
      "prompt_name": "tldr_en",
      "rougeL_fmeasure": 0.0429641799328658,
      "dataset_path": "GEM/wiki_lingua",
      "dataset_name": "en",
      "subset": null,
      "rougeL_fmeasure_stderr": 0.0014362645716637073
    },
    {
      "task_name": "GEM/wiki_lingua_en",
      "prompt_name": "tldr_en",
      "rougeLsum_precision": 0.049203521589357486,
      "dataset_path": "GEM/wiki_lingua",
      "dataset_name": "en",
      "subset": null,
      "rougeLsum_precision_stderr": 0.0018745212624263795
    },
    {
      "task_name": "GEM/wiki_lingua_en",
      "prompt_name": "tldr_en",
      "rougeLsum_recall": 0.07828616324493042,
      "dataset_path": "GEM/wiki_lingua",
      "dataset_name": "en",
      "subset": null,
      "rougeLsum_recall_stderr": 0.00270998239625019
    },
    {
      "task_name": "GEM/wiki_lingua_en",
      "prompt_name": "tldr_en",
      "rougeLsum_fmeasure": 0.052823275351703086,
      "dataset_path": "GEM/wiki_lingua",
      "dataset_name": "en",
      "subset": null,
      "rougeLsum_fmeasure_stderr": 0.0018085153746161442
    },
    {
      "task_name": "GEM/wiki_lingua_en",
      "prompt_name": "tldr_en",
      "bleu": 0.5159665881377578,
      "dataset_path": "GEM/wiki_lingua",
      "dataset_name": "en",
      "subset": null,
      "bleu_stderr": 0.0355848354666848
    }
  ],
  "config": {
    "model": "hf-causal",
    "model_args": "pretrained=/pfs/lustrep4/scratch/project_462000119/muennighoff/nov-2022-bettercom/lm1-4b2-84b-c4-repetitions/4b284b12bc4/transformers,use_accelerate=True,tokenizer=/pfs/lustrep4/scratch/project_462000119/muennighoff/nov-2022-bettercom/gpt2,dtype=bfloat16",
    "task_args": "",
    "num_fewshot": 4,
    "batch_size": 16,
    "device": "cuda",
    "use_cache": false,
    "limit": 3000,
    "bootstrap_iters": 10,
    "seed": 1234
  }
}