8b712b12b/evaluation/generation/slim.limited=3000.model=lm1-8b7-12b.task=GEM-web_nlg_en.templates=PALM_prompt.fewshot=3.batchsize=16.seed=1234.timestamp=2023-01-24T16:59:52.json

{
  "results": [
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "bleu": 0.32359241899151747,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "bleu_stderr": 0.02402979694298941
    },
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "rouge1_precision": 0.08608348405841157,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "rouge1_precision_stderr": 0.0030889568803165366
    },
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "rouge1_recall": 0.24581368650128066,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "rouge1_recall_stderr": 0.0042101313562470394
    },
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "rouge1_fmeasure": 0.10561184879910503,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "rouge1_fmeasure_stderr": 0.0025642659603447846
    },
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "rouge2_precision": 0.03877151013301891,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "rouge2_precision_stderr": 0.0018686193069556473
    },
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "rouge2_recall": 0.11616825070098583,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "rouge2_recall_stderr": 0.002832186555639384
    },
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "rouge2_fmeasure": 0.04745672201410211,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "rouge2_fmeasure_stderr": 0.0014569648711692605
    },
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "rougeL_precision": 0.07844420087825354,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "rougeL_precision_stderr": 0.002692441142953267
    },
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "rougeL_recall": 0.23451408639630753,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "rougeL_recall_stderr": 0.003961883955493253
    },
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "rougeL_fmeasure": 0.09789371482528919,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "rougeL_fmeasure_stderr": 0.00220103630816279
    },
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "rougeLsum_precision": 0.07915325102565371,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "rougeLsum_precision_stderr": 0.0027526142780926856
    },
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "rougeLsum_recall": 0.23412111056208834,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "rougeLsum_recall_stderr": 0.003946618325848973
    },
    {
      "task_name": "GEM/web_nlg_en",
      "prompt_name": "PALM_prompt",
      "rougeLsum_fmeasure": 0.09830813287691423,
      "dataset_path": "GEM/web_nlg",
      "dataset_name": "en",
      "subset": null,
      "rougeLsum_fmeasure_stderr": 0.0022462680004158276
    }
  ],
  "config": {
    "model": "hf-causal",
    "model_args": "pretrained=/pfs/lustrep4/scratch/project_462000119/muennighoff/nov-2022-bettercom/lm1-8b7-12b/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
  }
}