pl/polish_benchmarks-out5/results_trurl-2-13b-academic_8bits-5_polish_generate/results.json

{
  "results": {
    "polish_generate": {
      "exact_match,score-first": 0.6599270453361126,
      "exact_match_stderr,score-first": 0.08111767709880116,
      "alias": "polish_generate"
    },
    "polemo2_in": {
      "exact_match,score-first": 0.7091412742382271,
      "exact_match_stderr,score-first": 0.016913745750075915,
      "alias": " - polemo2_in"
    },
    "polemo2_out": {
      "exact_match,score-first": 0.6356275303643725,
      "exact_match_stderr,score-first": 0.02167457886979786,
      "alias": " - polemo2_out"
    },
    "polish_8tags_regex": {
      "exact_match,score-first": 0.7076852698993595,
      "exact_match_stderr,score-first": 0.006879470696168756,
      "alias": " - polish_8tags_regex"
    },
    "polish_belebele_regex": {
      "exact_match,score-first": 0.5888888888888889,
      "exact_match_stderr,score-first": 0.016410297014042757,
      "alias": " - polish_belebele_regex"
    },
    "polish_dyk_regex": {
      "exact_match,score-first": 0.7405247813411079,
      "exact_match_stderr,score-first": 0.013671657949578813,
      "alias": " - polish_dyk_regex"
    },
    "polish_ppc_regex": {
      "exact_match,score-first": 0.657,
      "exact_match_stderr,score-first": 0.015019206922356951,
      "alias": " - polish_ppc_regex"
    },
    "polish_psc_regex": {
      "exact_match,score-first": 0.4294990723562152,
      "exact_match_stderr,score-first": 0.015083476149096428,
      "alias": " - polish_psc_regex"
    }
  },
  "groups": {
    "polish_generate": {
      "exact_match,score-first": 0.6599270453361126,
      "exact_match_stderr,score-first": 0.08111767709880116,
      "alias": "polish_generate"
    }
  },
  "configs": {
    "polemo2_in": {
      "task": "polemo2_in",
      "group": [
        "polemo2"
      ],
      "dataset_path": "allegro/klej-polemo2-in",
      "training_split": "train",
      "validation_split": "validation",
      "test_split": "test",
      "doc_to_text": "Opinia: \"{{sentence}}\"\nOkreśl sentyment podanej opinii. Możliwe odpowiedzi:\nA - Neutralny\nB - Negatywny\nC - Pozytywny\nD - Niejednoznaczny\nPrawidłowa odpowiedź:",
      "doc_to_target": "{{{'__label__meta_zero': 'A', '__label__meta_minus_m': 'B', '__label__meta_plus_m': 'C', '__label__meta_amb': 'D'}.get(target)}}",
      "description": "",
      "target_delimiter": " ",
      "fewshot_delimiter": "\n\n",
      "num_fewshot": 5,
      "metric_list": [
        {
          "metric": "exact_match",
          "aggregation": "mean",
          "higher_is_better": true
        }
      ],
      "output_type": "generate_until",
      "generation_kwargs": {
        "until": [
          ".",
          ","
        ],
        "do_sample": false,
        "temperature": 0.0,
        "max_gen_toks": 50
      },
      "repeats": 1,
      "filter_list": [
        {
          "name": "score-first",
          "filter": [
            {
              "function": "regex",
              "regex_pattern": "(\\b[ABCD]\\b)"
            },
            {
              "function": "take_first"
            }
          ]
        }
      ],
      "should_decontaminate": true,
      "doc_to_decontamination_query": "{{sentence}}",
      "metadata": {
        "version": 1.0
      }
    },
    "polemo2_out": {
      "task": "polemo2_out",
      "group": [
        "polemo2"
      ],
      "dataset_path": "allegro/klej-polemo2-out",
      "training_split": "train",
      "validation_split": "validation",
      "test_split": "test",
      "doc_to_text": "Opinia: \"{{sentence}}\"\nOkreśl sentyment podanej opinii. Możliwe odpowiedzi:\nA - Neutralny\nB - Negatywny\nC - Pozytywny\nD - Niejednoznaczny\nPrawidłowa odpowiedź:",
      "doc_to_target": "{{{'__label__meta_zero': 'A', '__label__meta_minus_m': 'B', '__label__meta_plus_m': 'C', '__label__meta_amb': 'D'}.get(target)}}",
      "description": "",
      "target_delimiter": " ",
      "fewshot_delimiter": "\n\n",
      "num_fewshot": 5,
      "metric_list": [
        {
          "metric": "exact_match",
          "aggregation": "mean",
          "higher_is_better": true
        }
      ],
      "output_type": "generate_until",
      "generation_kwargs": {
        "until": [
          ".",
          ","
        ],
        "do_sample": false,
        "temperature": 0.0,
        "max_gen_toks": 50
      },
      "repeats": 1,
      "filter_list": [
        {
          "name": "score-first",
          "filter": [
            {
              "function": "regex",
              "regex_pattern": "(\\b[ABCD]\\b)"
            },
            {
              "function": "take_first"
            }
          ]
        }
      ],
      "should_decontaminate": true,
      "doc_to_decontamination_query": "{{sentence}}",
      "metadata": {
        "version": 1.0
      }
    },
    "polish_8tags_regex": {
      "task": "polish_8tags_regex",
      "dataset_path": "sdadas/8tags",
      "training_split": "train",
      "validation_split": "validation",
      "test_split": "test",
      "doc_to_text": "Tytuł: \"{{sentence}}\"\nPytanie: jaka kategoria najlepiej pasuje do podanego tytułu?\nMożliwe odpowiedzi:\nA - film\nB - historia\nC - jedzenie\nD - medycyna\nE - motoryzacja\nF - praca\nG - sport\nH - technologie\nPrawidłowa odpowiedź:",
      "doc_to_target": "{{{0: 'A', 1: 'B', 2: 'C', 3: 'D', 4: 'E', 5: 'F', 6: 'G', 7: 'H'}.get(label)}}",
      "description": "",
      "target_delimiter": " ",
      "fewshot_delimiter": "\n\n",
      "num_fewshot": 5,
      "metric_list": [
        {
          "metric": "exact_match",
          "aggregation": "mean",
          "higher_is_better": true
        }
      ],
      "output_type": "generate_until",
      "generation_kwargs": {
        "until": [
          ".",
          ","
        ],
        "do_sample": false,
        "temperature": 0.0,
        "max_gen_toks": 50
      },
      "repeats": 1,
      "filter_list": [
        {
          "name": "score-first",
          "filter": [
            {
              "function": "regex",
              "regex_pattern": "(\\b[ABCDEFGH]\\b)"
            },
            {
              "function": "take_first"
            }
          ]
        }
      ],
      "should_decontaminate": true,
      "doc_to_decontamination_query": "{{sentence}}"
    },
    "polish_belebele_regex": {
      "task": "polish_belebele_regex",
      "dataset_path": "facebook/belebele",
      "test_split": "pol_Latn",
      "doc_to_text": "Fragment: \"{{flores_passage}}\"\nPytanie: \"{{question}}\"\nMożliwe odpowiedzi:\nA - {{mc_answer1}}\nB - {{mc_answer2}}\nC - {{mc_answer3}}\nD - {{mc_answer4}}\nPrawidłowa odpowiedź:",
      "doc_to_target": "{{{0: 'A', 1: 'B', 2: 'C', 3: 'D'}.get(correct_answer_num|int - 1)}}",
      "description": "",
      "target_delimiter": " ",
      "fewshot_delimiter": "\n\n",
      "num_fewshot": 5,
      "metric_list": [
        {
          "metric": "exact_match",
          "aggregation": "mean",
          "higher_is_better": true
        }
      ],
      "output_type": "generate_until",
      "generation_kwargs": {
        "until": [
          ".",
          ","
        ],
        "do_sample": false,
        "temperature": 0.0,
        "max_gen_toks": 50
      },
      "repeats": 1,
      "filter_list": [
        {
          "name": "score-first",
          "filter": [
            {
              "function": "regex",
              "regex_pattern": "(\\b[ABCD]\\b)"
            },
            {
              "function": "take_first"
            }
          ]
        }
      ],
      "should_decontaminate": true,
      "doc_to_decontamination_query": "{{flores_passage}} {{question}} {{mc_answer1}} {{mc_answer2}} {{mc_answer3}} {{mc_answer4}}"
    },
    "polish_dyk_regex": {
      "task": "polish_dyk_regex",
      "dataset_path": "allegro/klej-dyk",
      "training_split": "train",
      "test_split": "test",
      "doc_to_text": "Pytanie: \"{{question}}\"\nSugerowana odpowiedź: \"{{answer}}\"\nCzy sugerowana odpowiedź na zadane pytanie jest poprawna? Możliwe opcje:\nA - brakuje sugerowanej odpowiedzi\nB - nie, sugerowana odpowiedź nie jest poprawna\nC - tak, sugerowana odpowiedź jest poprawna\nD - brakuje pytania\nPrawidłowa opcja:",
      "doc_to_target": "{{{0: 'A', 1: 'B', 2: 'C', 3: 'D'}.get(target|int + 1)}}",
      "description": "",
      "target_delimiter": " ",
      "fewshot_delimiter": "\n\n",
      "num_fewshot": 5,
      "metric_list": [
        {
          "metric": "exact_match",
          "aggregation": "mean",
          "higher_is_better": true
        }
      ],
      "output_type": "generate_until",
      "generation_kwargs": {
        "until": [
          ".",
          ","
        ],
        "do_sample": false,
        "temperature": 0.0,
        "max_gen_toks": 50
      },
      "repeats": 1,
      "filter_list": [
        {
          "name": "score-first",
          "filter": [
            {
              "function": "regex",
              "regex_pattern": "(\\b[ABCD]\\b)"
            },
            {
              "function": "take_first"
            }
          ]
        }
      ],
      "should_decontaminate": true,
      "doc_to_decontamination_query": "{{question}} {{answer}}"
    },
    "polish_ppc_regex": {
      "task": "polish_ppc_regex",
      "dataset_path": "sdadas/ppc",
      "training_split": "train",
      "validation_split": "validation",
      "test_split": "test",
      "doc_to_text": "Zdanie A: \"{{sentence_A}}\"\nZdanie B: \"{{sentence_B}}\"\nPytanie: jaka jest zależność między zdaniami A i B? Możliwe odpowiedzi:\nA - wszystkie odpowiedzi poprawne\nB - znaczą dokładnie to samo\nC - mają podobne znaczenie\nD - mają różne znaczenie\nPrawidłowa odpowiedź:",
      "doc_to_target": "{{{0: 'A', 1: 'B', 2: 'C', 3: 'D'}.get(label|int)}}",
      "description": "",
      "target_delimiter": " ",
      "fewshot_delimiter": "\n\n",
      "num_fewshot": 5,
      "metric_list": [
        {
          "metric": "exact_match",
          "aggregation": "mean",
          "higher_is_better": true
        }
      ],
      "output_type": "generate_until",
      "generation_kwargs": {
        "until": [
          ".",
          ","
        ],
        "do_sample": false,
        "temperature": 0.0,
        "max_gen_toks": 50
      },
      "repeats": 1,
      "filter_list": [
        {
          "name": "score-first",
          "filter": [
            {
              "function": "regex",
              "regex_pattern": "(\\b[ABCD]\\b)"
            },
            {
              "function": "take_first"
            }
          ]
        }
      ],
      "should_decontaminate": true,
      "doc_to_decontamination_query": "{{sentence_A}} {{sentence_B}}"
    },
    "polish_psc_regex": {
      "task": "polish_psc_regex",
      "dataset_path": "allegro/klej-psc",
      "training_split": "train",
      "test_split": "test",
      "doc_to_text": "Fragment 1: \"{{extract_text}}\"\nFragment 2: \"{{summary_text}}\"\nPytanie: jaka jest zależność między fragmentami 1 i 2?\nMożliwe odpowiedzi:\nA - wszystkie odpowiedzi poprawne\nB - dotyczą tego samego artykułu\nC - dotyczą różnych artykułów\nD - brak poprawnej odpowiedzi\nPrawidłowa odpowiedź:",
      "doc_to_target": "{{{0: 'A', 1: 'B', 2: 'C', 3: 'D'}.get(label|int + 1)}}",
      "description": "",
      "target_delimiter": " ",
      "fewshot_delimiter": "\n\n",
      "num_fewshot": 5,
      "metric_list": [
        {
          "metric": "exact_match",
          "aggregation": "mean",
          "higher_is_better": true
        }
      ],
      "output_type": "generate_until",
      "generation_kwargs": {
        "until": [
          ".",
          ","
        ],
        "do_sample": false,
        "temperature": 0.0,
        "max_gen_toks": 50
      },
      "repeats": 1,
      "filter_list": [
        {
          "name": "score-first",
          "filter": [
            {
              "function": "regex",
              "regex_pattern": "(\\b[ABCD]\\b)"
            },
            {
              "function": "take_first"
            }
          ]
        }
      ],
      "should_decontaminate": true,
      "doc_to_decontamination_query": "{{extract_text}} {{summary_text}}"
    }
  },
  "versions": {
    "polemo2_in": 1.0,
    "polemo2_out": 1.0,
    "polish_8tags_regex": "Yaml",
    "polish_belebele_regex": "Yaml",
    "polish_dyk_regex": "Yaml",
    "polish_generate": "N/A",
    "polish_ppc_regex": "Yaml",
    "polish_psc_regex": "Yaml"
  },
  "n-shot": {
    "polemo2_in": 5,
    "polemo2_out": 5,
    "polish_8tags_regex": 5,
    "polish_belebele_regex": 5,
    "polish_dyk_regex": 5,
    "polish_generate": 5,
    "polish_ppc_regex": 5,
    "polish_psc_regex": 5
  },
  "config": {
    "model": "hf",
    "model_args": "pretrained=Voicelab/trurl-2-13b-academic",
    "batch_size": "1",
    "batch_sizes": [],
    "device": "cuda:0",
    "use_cache": null,
    "limit": null,
    "bootstrap_iters": 100000,
    "gen_kwargs": null
  },
  "git_hash": null
}