results.py

"""MTEB Results"""

import json
import collections
import os
import datasets
import huggingface_hub

logger = datasets.logging.get_logger(__name__)


_CITATION = """@article{muennighoff2022mteb,
  doi = {10.48550/ARXIV.2210.07316},
  url = {https://arxiv.org/abs/2210.07316},
  author = {Muennighoff, Niklas and Tazi, Nouamane and Magne, Lo{\"\i}c and Reimers, Nils},
  title = {MTEB: Massive Text Embedding Benchmark},
  publisher = {arXiv},
  journal={arXiv preprint arXiv:2210.07316},  
  year = {2022}
}
"""

_DESCRIPTION = """Results on MTEB"""

#URL = "https://huggingface.co/datasets/pt-mteb/results/resolve/main/paths.json"
REPO_ID = "pt-mteb/results"
VERSION = datasets.Version("1.0.1")
EVAL_LANGS = ['af', 'afr-por', 'am', "amh", 'amh-por', 'ang-por', 'ar', 'ar-ar', 'ara-por', 'arq-por', 'arz-por', 'ast-por', 'awa-por', 'az', 'aze-por', 'bel-por', 'ben-por', 'ber-por', 'bn', 'bos-por', 'bre-por', 'bul-por', 'cat-por', 'cbk-por', 'ceb-por', 'ces-por', 'cha-por', 'cmn-por', 'cor-por', 'csb-por', 'cy', 'cym-por', 'da', 'dan-por', 'de', 'de-fr', 'de-pl', 'deu-por', 'dsb-por', 'dtp-por', 'el', 'ell-por', 'en', 'pt-ar', 'pt-de', 'pt-pt', 'pt-tr', 'por', 'epo-por', 'es', 'es-pt', 'es-es', 'es-it', 'est-por', 'eus-por', 'fa', 'fao-por', 'fi', 'fin-por', 'fr', 'fr-pt', 'fr-pl', 'fra', 'fra-por', 'fry-por', 'gla-por', 'gle-por', 'glg-por', 'gsw-por', 'hau', 'he', 'heb-por', 'hi', 'hin-por', 'hrv-por', 'hsb-por', 'hu', 'hun-por', 'hy', 'hye-por', 'ibo', 'id', 'ido-por', 'ile-por', 'ina-por', 'ind-por', 'is', 'isl-por', 'it', 'it-pt', 'ita-por', 'ja', 'jav-por', 'jpn-por', 'jv', 'ka', 'kab-por', 'kat-por', 'kaz-por', 'khm-por', 'km', 'kn', 'ko', 'ko-ko', 'kor-por', 'kur-por', 'kzj-por', 'lat-por', 'lfn-por', 'lit-por', 'lin', 'lug', 'lv', 'lvs-por', 'mal-por', 'mar-por', 'max-por', 'mhr-por', 'mkd-por', 'ml', 'mn', 'mon-por', 'ms', 'my', 'nb', 'nds-por', 'nl', 'nl-ptde-pt', 'nld-por', 'nno-por', 'nob-por', 'nov-por', 'oci-por', 'orm', 'orv-por', 'pam-por', 'pcm', 'pes-por', 'pl', 'pl-pt', 'pms-por', 'pol-por', 'por-por', 'pt', 'ro', 'ron-por', 'ru', 'run', 'rus-por', 'sl', 'slk-por', 'slv-por', 'spa-por', 'sna', 'som', 'sq', 'sqi-por', 'srp-por', 'sv', 'sw', 'swa', 'swe-por', 'swg-por', 'swh-por', 'ta', 'tam-por', 'tat-por', 'te', 'tel-por', 'tgl-por', 'th', 'tha-por', 'tir', 'tl', 'tr', 'tuk-por', 'tur-por', 'tzl-por', 'uig-por', 'ukr-por', 'ur', 'urd-por', 'uzb-por', 'vi', 'vie-por', 'war-por', 'wuu-por', 'xho', 'xho-por', 'yid-por', 'yor', 'yue-por', 'zh', 'zh-CN', 'zh-TW', 'zh-pt', 'zsm-por', "eng_Latn-por_Latn","spa_Latn-por_Latn","fra_Latn-por_Latn","ita_Latn-por_Latn","deu_Latn-por_Latn","jpn_Jpan-por_Latn","kor_Hang-por_Latn","rus_Cyrl-por_Latn","arb_Arab-por_Latn","zho_Hant-por_Latn","zho_Hans-por_Latn","pol_Latn-por_Latn","swe_Latn-por_Latn"]

SKIP_KEYS = ["std", "evaluation_time", "main_score", "threshold"]

# Use "train" split instead
TRAIN_SPLIT = ["DanishPoliticalCommentsClassification"]
# Use "validation" split instead
VALIDATION_SPLIT = ["AFQMC", "Cmnli", "IFlyTek", "TNews", "MSMARCO", "MSMARCO-PL", "MultilingualSentiment", "Ocnli"]
# Use "dev" split instead
DEV_SPLIT = ["CmedqaRetrieval", "CovidRetrieval", "DuRetrieval", "EcomRetrieval", "MedicalRetrieval", "MMarcoReranking", "MMarcoRetrieval", "MSMARCO", "MSMARCO-PL", "T2Reranking", "T2Retrieval", "VideoRetrieval", "FloresBitextMining"]
# Use "test.full" split
TESTFULL_SPLIT = ["OpusparcusPC"]

# Needs to be run whenever new files are added
def get_paths():
    import collections, json, os
    files = collections.defaultdict(list)
    for base in os.listdir("results"):
        if not os.path.isdir(os.path.join("results", base)):
            continue
        results_base_dir = os.path.join("results", base)
        result_dirs = []
        for d in os.listdir(results_base_dir):
            current_path = os.path.join(results_base_dir, d)
            added_root = False
            if os.path.isdir(current_path):
                result_dirs.append((os.path.join(base,d), current_path))
            elif current_path.endswith('.json') and not added_root:
                result_dirs.append((base, results_base_dir))
                added_root = True
        for model_dir, results_model_dir in result_dirs:
            for res_file in os.listdir(results_model_dir):
                if res_file.endswith(".json"):
                    results_model_file = os.path.join(results_model_dir, res_file)
                    files[model_dir].append(results_model_file)
    with open("paths.json", "w") as f:
        json.dump(files, f)
    return files

#paths = get_paths()
#paths = json.loads(requests.get(URL).content.decode('utf8'))

dataset_info = huggingface_hub.dataset_info(REPO_ID)
filepaths = [r.rfilename for r in dataset_info.siblings]
paths = collections.defaultdict(list)
for filepath in filepaths:
    if 'ipynb' in filepath:
        continue
    if not filepath.startswith('results/') or not filepath.endswith('.json'):
        continue
    model_name = os.path.dirname(filepath)
    model_name = model_name[len('results/'):]
    paths[model_name].append(filepath)

MODELS = list(paths.keys())

class MTEBConfig(datasets.BuilderConfig):

    def __init__(self,
                complete_name=None,
                *args,
                **kwargs):
        super().__init__(*args, **kwargs)
        self.complete_name = complete_name

class MTEBResults(datasets.GeneratorBasedBuilder):
    """MTEBResults"""

    BUILDER_CONFIGS = [
        MTEBConfig(
            name=model.replace('/', '__') if '/' in model else model,
            description=f"{model} MTEB results",
            version=VERSION,
            complete_name=model,
        )
        for model in MODELS
    ]

    def _info(self):
        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            features=datasets.Features(
                {
                    "mteb_dataset_name": datasets.Value("string"),
                    "eval_language": datasets.Value("string"),
                    "metric": datasets.Value("string"),
                    "score": datasets.Value("float"),
                }
            ),
            supervised_keys=None,
            citation=_CITATION,
        )

    def _split_generators(self, dl_manager):
        global paths
        #path_file = dl_manager.download_and_extract(URL)
        #with open(path_file) as f:
        #    files = json.load(f)
        files = paths
        
        downloaded_files = dl_manager.download_and_extract(files[self.config.complete_name])
        return [
            datasets.SplitGenerator(
                name=datasets.Split.TEST,
                gen_kwargs={'filepath': downloaded_files}
            )
        ]

    def _generate_examples(self, filepath):
        """This function returns the examples in the raw (text) form."""
        logger.info(f"Generating examples from {filepath}")

        out = []

        for path in filepath:
            with open(path, encoding="utf-8") as f:
                res_dict = json.load(f)
                ds_name = res_dict["mteb_dataset_name"]
                split = "test"
                if (ds_name in TRAIN_SPLIT) and ("train" in res_dict):
                    split = "train"
                elif (ds_name in VALIDATION_SPLIT) and ("validation" in res_dict):
                    split = "validation"
                elif (ds_name in DEV_SPLIT) and ("dev" in res_dict):
                    split = "dev"
                elif (ds_name in TESTFULL_SPLIT) and ("test.full" in res_dict):
                    split = "test.full"
                elif "test" not in res_dict:
                    print(f"Skipping {ds_name} as split {split} not present.")
                    continue
                res_dict = res_dict.get(split)
                is_multilingual = any(x in res_dict for x in EVAL_LANGS)
                langs = res_dict.keys() if is_multilingual else ["pt"]
                for lang in langs:
                    if lang in SKIP_KEYS: continue
                    test_result_lang = res_dict.get(lang) if is_multilingual else res_dict
                    for metric, score in test_result_lang.items():
                        if not isinstance(score, dict):
                            score = {metric: score}
                        for sub_metric, sub_score in score.items():
                            if any(x in sub_metric for x in SKIP_KEYS): continue
                            if isinstance(sub_score, dict) or isinstance(sub_score, list): continue
                            out.append({
                                "mteb_dataset_name": ds_name,
                                "eval_language": lang if is_multilingual else "",
                                "metric": f"{metric}_{sub_metric}" if metric != sub_metric else metric,
                                "score": sub_score * 100,
                            })
        for idx, row in enumerate(sorted(out, key=lambda x: x["mteb_dataset_name"])):
            yield idx, row