8b7178b58b/evaluation/generation/slim.8b7178b58b_GEM-web_nlg_en_PALM_prompt_3.json

{
  "results": [
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "bleu": 0.6611202603466564,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "bleu_stderr": 0.026292760196011896
    },
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "rouge1_precision": 0.07720993388700956,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "rouge1_precision_stderr": 0.0012101497257156794
    },
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "rouge1_recall": 0.4133639192496906,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "rouge1_recall_stderr": 0.005411628648726416
    },
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "rouge1_fmeasure": 0.12270682587650009,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "rouge1_fmeasure_stderr": 0.001689827532535868
    },
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "rouge2_precision": 0.03655832324841317,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "rouge2_precision_stderr": 0.0007506453443122757
    },
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "rouge2_recall": 0.21585685095591742,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "rouge2_recall_stderr": 0.004093725567157412
    },
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "rouge2_fmeasure": 0.05861591086080908,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "rouge2_fmeasure_stderr": 0.0010907460831452867
    },
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "rougeL_precision": 0.07151482183987531,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "rougeL_precision_stderr": 0.001086199341044554
    },
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "rougeL_recall": 0.3823649104455851,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "rougeL_recall_stderr": 0.004813722726129101
    },
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "rougeL_fmeasure": 0.11361332097070079,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "rougeL_fmeasure_stderr": 0.0015022331310385422
    },
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "rougeLsum_precision": 0.07350481492884033,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "rougeLsum_precision_stderr": 0.0011485603851221294
    },
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "rougeLsum_recall": 0.39303289732256513,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "rougeLsum_recall_stderr": 0.005032921309423209
    },
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "rougeLsum_fmeasure": 0.11674347677370367,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "rougeLsum_fmeasure_stderr": 0.0015911049077229342
    }
  ],
  "config": {
    "model": "hf-causal",
    "model_args": "pretrained=/pfs/lustrep4/scratch/project_462000119/muennighoff/nov-2022-bettercom/lm1-8b7-178b-c4-repetitions/8b7178b58b/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
  }
}