{ "reference": { "reference_perplexity": 24.01962701026171, "samples": 1000 }, "greedy": { "name": "greedy", "kind": "greedy", "samples": 1000, "generation_perplexity": 2.1845094968449197, "token_match_rate": 0.017536830158415048, "first_sentence_exact_match": 0.0, "mean_generated_tokens": 63.987, "total_seconds": 30.55351047298609, "seconds_per_sample": 0.03055351047298609, "reference_perplexity": 24.01962701026171, "bleu": 0.4011543259943025, "rouge_l": 0.10133870138074608, "repeat_rate": 0.991, "distinct_2": 0.37804763470662844 }, "beam-1": { "name": "beam-1", "kind": "beam", "width": 1, "samples": 1000, "generation_perplexity": 2.184142097873099, "token_match_rate": 0.01747016349174838, "first_sentence_exact_match": 0.0, "mean_generated_tokens": 63.987, "total_seconds": 748.7766567100189, "seconds_per_sample": 0.7487766567100189, "reference_perplexity": 24.01962701026171, "bleu": 0.4017249195346004, "rouge_l": 0.10134016059568843, "repeat_rate": 0.991, "distinct_2": 0.37787066171981404 }, "beam-2": { "name": "beam-2", "kind": "beam", "width": 2, "samples": 1000, "generation_perplexity": 1.9418890830063311, "token_match_rate": 0.012793069344256506, "first_sentence_exact_match": 0.0, "mean_generated_tokens": 63.82, "total_seconds": 743.5340949810707, "seconds_per_sample": 0.7435340949810707, "reference_perplexity": 24.01962701026171, "bleu": 0.26636636550312737, "rouge_l": 0.08545188908177155, "repeat_rate": 0.997, "distinct_2": 0.35923942153186933 }, "beam-4": { "name": "beam-4", "kind": "beam", "width": 4, "samples": 1000, "generation_perplexity": 1.7872872389121282, "token_match_rate": 0.013918423162327762, "first_sentence_exact_match": 0.0, "mean_generated_tokens": 63.519, "total_seconds": 745.1537463600471, "seconds_per_sample": 0.745153746360047, "reference_perplexity": 24.01962701026171, "bleu": 0.21235683373778194, "rouge_l": 0.07549435572208057, "repeat_rate": 0.997, "distinct_2": 0.3154023728995544 }, "top-k-5": { "name": "top-k-5", "kind": "top_k", "k": 5, "temperature": 1.0, "samples": 1000, "generation_perplexity": 5.840530230478968, "token_match_rate": 0.01660183385003512, "first_sentence_exact_match": 0.0, "mean_generated_tokens": 63.671, "total_seconds": 29.016556230009883, "seconds_per_sample": 0.02901655623000988, "reference_perplexity": 24.01962701026171, "bleu": 0.4880453760353093, "rouge_l": 0.1236210191578215, "repeat_rate": 0.393, "distinct_2": 0.8830857095559326 }, "top-k-20": { "name": "top-k-20", "kind": "top_k", "k": 20, "temperature": 1.0, "samples": 1000, "generation_perplexity": 10.019873678669159, "token_match_rate": 0.015090541951915377, "first_sentence_exact_match": 0.0, "mean_generated_tokens": 63.223, "total_seconds": 28.801222771980974, "seconds_per_sample": 0.028801222771980974, "reference_perplexity": 24.01962701026171, "bleu": 0.4856051512173579, "rouge_l": 0.11775486793844488, "repeat_rate": 0.115, "distinct_2": 0.9596821600145947 }, "top-k-50": { "name": "top-k-50", "kind": "top_k", "k": 50, "temperature": 1.0, "samples": 1000, "generation_perplexity": 13.80108373815267, "token_match_rate": 0.013518684208311345, "first_sentence_exact_match": 0.0, "mean_generated_tokens": 63.13, "total_seconds": 27.95822200999828, "seconds_per_sample": 0.02795822200999828, "reference_perplexity": 24.01962701026171, "bleu": 0.46471149496907277, "rouge_l": 0.11558306869222004, "repeat_rate": 0.053, "distinct_2": 0.9742885973763875 }, "top-p-0.5": { "name": "top-p-0.5", "kind": "top_p", "p": 0.5, "temperature": 1.0, "samples": 1000, "generation_perplexity": 5.867983666604579, "token_match_rate": 0.016987026890306695, "first_sentence_exact_match": 0.0, "mean_generated_tokens": 64.0, "total_seconds": 31.004208463971736, "seconds_per_sample": 0.031004208463971734, "reference_perplexity": 24.01962701026171, "bleu": 0.6045125923332172, "rouge_l": 0.11670411529976443, "repeat_rate": 0.384, "distinct_2": 0.8828828828828829 }, "top-p-0.9": { "name": "top-p-0.9", "kind": "top_p", "p": 0.9, "temperature": 1.0, "samples": 1000, "generation_perplexity": 24.27615581913303, "token_match_rate": 0.012577092470312138, "first_sentence_exact_match": 0.0, "mean_generated_tokens": 62.755, "total_seconds": 30.671832652995363, "seconds_per_sample": 0.030671832652995362, "reference_perplexity": 24.01962701026171, "bleu": 0.21955334969040827, "rouge_l": 0.10558525109265285, "repeat_rate": 0.021, "distinct_2": 0.9854764735858672 }, "top-p-0.95": { "name": "top-p-0.95", "kind": "top_p", "p": 0.95, "temperature": 1.0, "samples": 1000, "generation_perplexity": 34.31498216399583, "token_match_rate": 0.012355755202568324, "first_sentence_exact_match": 0.0, "mean_generated_tokens": 62.609, "total_seconds": 29.93822276096762, "seconds_per_sample": 0.02993822276096762, "reference_perplexity": 24.01962701026171, "bleu": 0.23364194724538673, "rouge_l": 0.1033430667860992, "repeat_rate": 0.022, "distinct_2": 0.987771768176668 } }