8b7178b58b/evaluation/generation/slim.8b7178b58b_GEM-web_nlg_en_PALM_prompt_2.json

{
  "results": [
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "bleu": 0.6345753558723145,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "bleu_stderr": 0.022376664491002615
    },
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "rouge1_precision": 0.07795669324486854,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "rouge1_precision_stderr": 0.00119684640928671
    },
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "rouge1_recall": 0.41803228663523345,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "rouge1_recall_stderr": 0.0053864905798921125
    },
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "rouge1_fmeasure": 0.12360575844131859,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "rouge1_fmeasure_stderr": 0.0016749434032436248
    },
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "rouge2_precision": 0.03697168943889819,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "rouge2_precision_stderr": 0.0007524504443837138
    },
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "rouge2_recall": 0.21717607459640267,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "rouge2_recall_stderr": 0.004101036268432045
    },
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "rouge2_fmeasure": 0.059094553406324295,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "rouge2_fmeasure_stderr": 0.0010986707292730262
    },
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "rougeL_precision": 0.07270569406411607,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "rougeL_precision_stderr": 0.0010983482762338668
    },
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "rougeL_recall": 0.38609502456284567,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "rougeL_recall_stderr": 0.004797386037833962
    },
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "rougeL_fmeasure": 0.1151080347627734,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "rougeL_fmeasure_stderr": 0.0015232281204089519
    },
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "rougeLsum_precision": 0.07442415306126304,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "rougeLsum_precision_stderr": 0.0011448850348819214
    },
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "rougeLsum_recall": 0.396883636105679,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "rougeLsum_recall_stderr": 0.0049884922683259055
    },
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "rougeLsum_fmeasure": 0.11783674396217436,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "rougeLsum_fmeasure_stderr": 0.0015879949279193565
    }
  ],
  "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": 2,
    "batch_size": 16,
    "device": "cuda",
    "use_cache": false,
    "limit": 3000,
    "bootstrap_iters": 10,
    "seed": 1234
  }
}