{"results": [{"task_name": "mlsum_es", "prompt_name": "summarise_this_in_es_few_sentences", "bleu": 2.2245794650879462, "fixed_answer_choice_list": null, "dataset_path": "GEM/mlsum", "dataset_name": "es", "subset": "", "prompt_id": "5e644239-d989-4531-b2ff-44b0e4310df6", "prompt_jinja": "{{text}}\n===\nGiven the above document, write few sentences in Spanish to summarize: ||| {{target}}", "prompt_original_task": true, "comment": "", "bleu_stderr": 0.07743700029169612}, {"task_name": "mlsum_es", "prompt_name": "summarise_this_in_es_few_sentences", "rouge1_precision": 0.18419910608261986, "fixed_answer_choice_list": null, "dataset_path": "GEM/mlsum", "dataset_name": "es", "subset": "", "prompt_id": "5e644239-d989-4531-b2ff-44b0e4310df6", "prompt_jinja": "{{text}}\n===\nGiven the above document, write few sentences in Spanish to summarize: ||| {{target}}", "prompt_original_task": true, "comment": "", "rouge1_precision_stderr": 0.002301564923577535}, {"task_name": "mlsum_es", "prompt_name": "summarise_this_in_es_few_sentences", "rouge1_recall": 0.33528109600140793, "fixed_answer_choice_list": null, "dataset_path": "GEM/mlsum", "dataset_name": "es", "subset": "", "prompt_id": "5e644239-d989-4531-b2ff-44b0e4310df6", "prompt_jinja": "{{text}}\n===\nGiven the above document, write few sentences in Spanish to summarize: ||| {{target}}", "prompt_original_task": true, "comment": "", "rouge1_recall_stderr": 0.0034321373331462294}, {"task_name": "mlsum_es", "prompt_name": "summarise_this_in_es_few_sentences", "rouge1_fmeasure": 0.21126423815884174, "fixed_answer_choice_list": null, "dataset_path": "GEM/mlsum", "dataset_name": "es", "subset": "", "prompt_id": "5e644239-d989-4531-b2ff-44b0e4310df6", "prompt_jinja": "{{text}}\n===\nGiven the above document, write few sentences in Spanish to summarize: ||| {{target}}", "prompt_original_task": true, "comment": "", "rouge1_fmeasure_stderr": 0.0021771710222460634}, {"task_name": "mlsum_es", "prompt_name": "summarise_this_in_es_few_sentences", "rouge2_precision": 0.051112897675373886, "fixed_answer_choice_list": null, "dataset_path": "GEM/mlsum", "dataset_name": "es", "subset": "", "prompt_id": "5e644239-d989-4531-b2ff-44b0e4310df6", "prompt_jinja": "{{text}}\n===\nGiven the above document, write few sentences in Spanish to summarize: ||| {{target}}", "prompt_original_task": true, "comment": "", "rouge2_precision_stderr": 0.0014393881241720322}, {"task_name": "mlsum_es", "prompt_name": "summarise_this_in_es_few_sentences", "rouge2_recall": 0.09793074579590116, "fixed_answer_choice_list": null, "dataset_path": "GEM/mlsum", "dataset_name": "es", "subset": "", "prompt_id": "5e644239-d989-4531-b2ff-44b0e4310df6", "prompt_jinja": "{{text}}\n===\nGiven the above document, write few sentences in Spanish to summarize: ||| {{target}}", "prompt_original_task": true, "comment": "", "rouge2_recall_stderr": 0.0024100392963833633}, {"task_name": "mlsum_es", "prompt_name": "summarise_this_in_es_few_sentences", "rouge2_fmeasure": 0.05913033007358818, "fixed_answer_choice_list": null, "dataset_path": "GEM/mlsum", "dataset_name": "es", "subset": "", "prompt_id": "5e644239-d989-4531-b2ff-44b0e4310df6", "prompt_jinja": "{{text}}\n===\nGiven the above document, write few sentences in Spanish to summarize: ||| {{target}}", "prompt_original_task": true, "comment": "", "rouge2_fmeasure_stderr": 0.0014890254374386052}, {"task_name": "mlsum_es", "prompt_name": "summarise_this_in_es_few_sentences", "rougeL_precision": 0.13714024915254835, "fixed_answer_choice_list": null, "dataset_path": "GEM/mlsum", "dataset_name": "es", "subset": "", "prompt_id": "5e644239-d989-4531-b2ff-44b0e4310df6", "prompt_jinja": "{{text}}\n===\nGiven the above document, write few sentences in Spanish to summarize: ||| {{target}}", "prompt_original_task": true, "comment": "", "rougeL_precision_stderr": 0.0018260510947169805}, {"task_name": "mlsum_es", "prompt_name": "summarise_this_in_es_few_sentences", "rougeL_recall": 0.2521886801730905, "fixed_answer_choice_list": null, "dataset_path": "GEM/mlsum", "dataset_name": "es", "subset": "", "prompt_id": "5e644239-d989-4531-b2ff-44b0e4310df6", "prompt_jinja": "{{text}}\n===\nGiven the above document, write few sentences in Spanish to summarize: ||| {{target}}", "prompt_original_task": true, "comment": "", "rougeL_recall_stderr": 0.0027817174072391373}, {"task_name": "mlsum_es", "prompt_name": "summarise_this_in_es_few_sentences", "rougeL_fmeasure": 0.15711042852214044, "fixed_answer_choice_list": null, "dataset_path": "GEM/mlsum", "dataset_name": "es", "subset": "", "prompt_id": "5e644239-d989-4531-b2ff-44b0e4310df6", "prompt_jinja": "{{text}}\n===\nGiven the above document, write few sentences in Spanish to summarize: ||| {{target}}", "prompt_original_task": true, "comment": "", "rougeL_fmeasure_stderr": 0.0017251643310554304}, {"task_name": "mlsum_es", "prompt_name": "summarise_this_in_es_few_sentences", "rougeLsum_precision": 0.14518990658432604, "fixed_answer_choice_list": null, "dataset_path": "GEM/mlsum", "dataset_name": "es", "subset": "", "prompt_id": "5e644239-d989-4531-b2ff-44b0e4310df6", "prompt_jinja": "{{text}}\n===\nGiven the above document, write few sentences in Spanish to summarize: ||| {{target}}", "prompt_original_task": true, "comment": "", "rougeLsum_precision_stderr": 0.001900911801455617}, {"task_name": "mlsum_es", "prompt_name": "summarise_this_in_es_few_sentences", "rougeLsum_recall": 0.26906405645015485, "fixed_answer_choice_list": null, "dataset_path": "GEM/mlsum", "dataset_name": "es", "subset": "", "prompt_id": "5e644239-d989-4531-b2ff-44b0e4310df6", "prompt_jinja": "{{text}}\n===\nGiven the above document, write few sentences in Spanish to summarize: ||| {{target}}", "prompt_original_task": true, "comment": "", "rougeLsum_recall_stderr": 0.0030454611701701234}, {"task_name": "mlsum_es", "prompt_name": "summarise_this_in_es_few_sentences", "rougeLsum_fmeasure": 0.1670307582655529, "fixed_answer_choice_list": null, "dataset_path": "GEM/mlsum", "dataset_name": "es", "subset": "", "prompt_id": "5e644239-d989-4531-b2ff-44b0e4310df6", "prompt_jinja": "{{text}}\n===\nGiven the above document, write few sentences in Spanish to summarize: ||| {{target}}", "prompt_original_task": true, "comment": "", "rougeLsum_fmeasure_stderr": 0.0018288687884684008}], "config": {"model": "hf-causal", "model_args": "pretrained=/gpfsscratch/rech/six/commun/commun/experiments/muennighoff/bloomckpt/6b3t0/p31lossseqglobal_step1000,use_accelerate=True,tokenizer=/gpfsscratch/rech/six/commun/commun/experiments/muennighoff/bloomckpt/6b3t0/p31lossseqglobal_step1000,dtype=float16", "num_fewshot": 0, "batch_size": 16, "device": "cuda", "use_cache": false, "limit": 3000, "bootstrap_iters": 10, "seed": 1234}}