4b284b28bc4/evaluation/generation/slim.4b284b28bc4_GEM-web_nlg_en_PALM_prompt_3.json

{
  "results": [
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "bleu": 0.6240971401779115,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "bleu_stderr": 0.03840020245332954
    },
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "rouge1_precision": 0.07946921950052364,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "rouge1_precision_stderr": 0.0015734983149428565
    },
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "rouge1_recall": 0.41443709705557313,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "rouge1_recall_stderr": 0.0052063482293464285
    },
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "rouge1_fmeasure": 0.12413425365513392,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "rouge1_fmeasure_stderr": 0.0018801730074252724
    },
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "rouge2_precision": 0.03727374629297188,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "rouge2_precision_stderr": 0.0010068745541143666
    },
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "rouge2_recall": 0.20933728572588564,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "rouge2_recall_stderr": 0.0039729036741121115
    },
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "rouge2_fmeasure": 0.05836966723015618,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "rouge2_fmeasure_stderr": 0.0012404189838209753
    },
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "rougeL_precision": 0.07340984582992345,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "rougeL_precision_stderr": 0.0014110667213403662
    },
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "rougeL_recall": 0.3803567082991153,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "rougeL_recall_stderr": 0.004624202286888448
    },
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "rougeL_fmeasure": 0.11463438747216832,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "rougeL_fmeasure_stderr": 0.0016925693943895671
    },
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "rougeLsum_precision": 0.07570628299965847,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "rougeLsum_precision_stderr": 0.0015124490142500056
    },
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "rougeLsum_recall": 0.39213432148715394,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "rougeLsum_recall_stderr": 0.004828378800565317
    },
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "rougeLsum_fmeasure": 0.11802649327609507,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "rougeLsum_fmeasure_stderr": 0.0017907209912884955
    }
  ],
  "config": {
    "model": "hf-causal",
    "model_args": "pretrained=/pfs/lustrep4/scratch/project_462000119/muennighoff/nov-2022-bettercom/lm1-4b2-84b-c4-repetitions/4b284b28bc4/transformers,use_accelerate=True,tokenizer=/pfs/lustrep4/scratch/project_462000119/muennighoff/nov-2022-bettercom/gpt2,dtype=bfloat16",
    "task_args": "",
    "num_fewshot": 3,
    "batch_size": 16,
    "device": "cuda",
    "use_cache": false,
    "limit": 3000,
    "bootstrap_iters": 10,
    "seed": 1234
  }
}