8b7178b25b/evaluation/generation/slim.8b7178b25b_GEM-web_nlg_en_PALM_prompt_1.json

{
  "results": [
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "bleu": 0.4594356029919803,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "bleu_stderr": 0.025125558903473632
    },
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "rouge1_precision": 0.07027188536501679,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "rouge1_precision_stderr": 0.0012558026173472138
    },
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "rouge1_recall": 0.3450765288308058,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "rouge1_recall_stderr": 0.005055990718374951
    },
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "rouge1_fmeasure": 0.10986827089930207,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "rouge1_fmeasure_stderr": 0.0017427056384300328
    },
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "rouge2_precision": 0.03347891417115085,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "rouge2_precision_stderr": 0.0007753123397897254
    },
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "rouge2_recall": 0.1749435024767411,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "rouge2_recall_stderr": 0.0036147215578007344
    },
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "rouge2_fmeasure": 0.052665704185835084,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "rouge2_fmeasure_stderr": 0.001116219363919427
    },
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "rougeL_precision": 0.06659999496172114,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "rougeL_precision_stderr": 0.001145948530216064
    },
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "rougeL_recall": 0.32753689197885233,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "rougeL_recall_stderr": 0.004724892115185279
    },
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "rougeL_fmeasure": 0.10429693361721804,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "rougeL_fmeasure_stderr": 0.0016079944655777501
    },
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "rougeLsum_precision": 0.06687788385741289,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "rougeLsum_precision_stderr": 0.0011787909083336645
    },
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "rougeLsum_recall": 0.32738351484636163,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "rougeLsum_recall_stderr": 0.004663465991453483
    },
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "rougeLsum_fmeasure": 0.10453836123127251,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "rougeLsum_fmeasure_stderr": 0.0016341468252465234
    }
  ],
  "config": {
    "model": "hf-causal",
    "model_args": "pretrained=/pfs/lustrep4/scratch/project_462000119/muennighoff/nov-2022-bettercom/lm1-8b7-178b-c4-repetitions/8b7178b25b/transformers,use_accelerate=True,tokenizer=/pfs/lustrep4/scratch/project_462000119/muennighoff/nov-2022-bettercom/gpt2,dtype=bfloat16",
    "task_args": "",
    "num_fewshot": 1,
    "batch_size": 16,
    "device": "cuda",
    "use_cache": false,
    "limit": 3000,
    "bootstrap_iters": 10,
    "seed": 1234
  }
}