MTEB-lite: named reference models (models.csv), per-task calibration (calibration.csv), eng (v2) + eng-super (v3 selection with 46 reference models)
ac198e3 verified | {"task": "SICK-R", "type": "STS", "main_score": "cosine_spearman", "eval_split": "test", "hf_subsets": ["default"], "dataset": {"path": "mteb/sickr-sts", "revision": "20a6d6f312dd54037fe07a32d58e5e168867909d"}, "K": 20, "ids": ["s1494", "s1736", "s2135", "s2681", "s2966", "s3401", "s3662", "s3836", "s4873", "s5102", "s5677", "s6046", "s6201", "s6325", "s6523", "s7228", "s8266", "s9163", "s9556", "s9743"], "estimator": "set_metric", "calibration": {"a": 0.5576755642279938, "b": 0.4165366206846191, "n_fit": 46, "fit_r": 0.5888528441357828, "fit_resid_sd": 0.04027790218604393}, "item_key": "cos", "benchmark": "MTEB(eng, v2)"} |