gbert-large_deprel

This model is a fine-tuned version of deepset/gbert-large on the universal_dependencies dataset. It achieves the following results on the evaluation set:

• Loss: 0.5226
• : {'precision': 0.9634146341463414, 'recall': 0.9693251533742331, 'f1': 0.966360856269113, 'number': 163}
• Arataxis: {'precision': 0.28, 'recall': 0.2413793103448276, 'f1': 0.25925925925925924, 'number': 29}
• Ark: {'precision': 0.8518518518518519, 'recall': 0.8385416666666666, 'f1': 0.8451443569553806, 'number': 192}
• Ase: {'precision': 0.9595864661654135, 'recall': 0.9686907020872866, 'f1': 0.964117091595845, 'number': 1054}
• Bj: {'precision': 0.9388185654008439, 'recall': 0.8829365079365079, 'f1': 0.9100204498977505, 'number': 504}
• Bl: {'precision': 0.8804841149773072, 'recall': 0.8609467455621301, 'f1': 0.8706058339566194, 'number': 676}
• C: {'precision': 0.9455958549222798, 'recall': 0.9102244389027432, 'f1': 0.9275730622617535, 'number': 401}
• Cl: {'precision': 0.7558139534883721, 'recall': 0.6770833333333334, 'f1': 0.7142857142857142, 'number': 96}
• Comp: {'precision': 0.7674418604651163, 'recall': 0.7746478873239436, 'f1': 0.7710280373831776, 'number': 213}
• Dvcl: {'precision': 0.7922077922077922, 'recall': 0.7625, 'f1': 0.7770700636942675, 'number': 80}
• Dvmod: {'precision': 0.9073001158748552, 'recall': 0.903114186851211, 'f1': 0.9052023121387283, 'number': 867}
• Ep: {'precision': 0.6176470588235294, 'recall': 0.23863636363636365, 'f1': 0.3442622950819672, 'number': 88}
• Et: {'precision': 0.9549745824255628, 'recall': 0.9711964549483013, 'f1': 0.9630172098132551, 'number': 1354}
• Et:poss: {'precision': 0.9302325581395349, 'recall': 0.9448818897637795, 'f1': 0.9375, 'number': 127}
• Ixed: {'precision': 0.42857142857142855, 'recall': 0.2727272727272727, 'f1': 0.33333333333333326, 'number': 11}
• Lat: {'precision': 0.7272727272727273, 'recall': 0.8188976377952756, 'f1': 0.7703703703703703, 'number': 127}
• Mod: {'precision': 0.8328474246841594, 'recall': 0.8544366899302094, 'f1': 0.8435039370078741, 'number': 1003}
• Mod:poss: {'precision': 0.0, 'recall': 0.0, 'f1': 0.0, 'number': 0}
• Obj: {'precision': 0.9552238805970149, 'recall': 0.9142857142857143, 'f1': 0.9343065693430657, 'number': 70}
• Ocative: {'precision': 0.0, 'recall': 0.0, 'f1': 0.0, 'number': 2}
• Ompound: {'precision': 0.8111111111111111, 'recall': 0.5983606557377049, 'f1': 0.6886792452830189, 'number': 122}
• Ompound:prt: {'precision': 0.9078947368421053, 'recall': 0.8961038961038961, 'f1': 0.9019607843137255, 'number': 77}
• Onj: {'precision': 0.8546255506607929, 'recall': 0.8471615720524017, 'f1': 0.850877192982456, 'number': 458}
• Oot: {'precision': 0.9351620947630923, 'recall': 0.9398496240601504, 'f1': 0.9375, 'number': 798}
• Op: {'precision': 0.8957345971563981, 'recall': 0.9264705882352942, 'f1': 0.9108433734939759, 'number': 204}
• Ppos: {'precision': 0.7142857142857143, 'recall': 0.7851239669421488, 'f1': 0.7480314960629922, 'number': 121}
• Subj: {'precision': 0.9198355601233299, 'recall': 0.9049544994944388, 'f1': 0.9123343527013253, 'number': 989}
• Subj:pass: {'precision': 0.8666666666666667, 'recall': 0.9176470588235294, 'f1': 0.8914285714285715, 'number': 85}
• Ummod: {'precision': 0.9126984126984127, 'recall': 0.8646616541353384, 'f1': 0.888030888030888, 'number': 133}
• Unct: {'precision': 0.9735142118863049, 'recall': 0.9592616168045831, 'f1': 0.9663353638986855, 'number': 1571}
• Ux: {'precision': 0.9683544303797469, 'recall': 0.9216867469879518, 'f1': 0.9444444444444444, 'number': 332}
• Ux:pass: {'precision': 0.8653846153846154, 'recall': 0.9278350515463918, 'f1': 0.8955223880597015, 'number': 97}
• Xpl: {'precision': 0.37037037037037035, 'recall': 0.7692307692307693, 'f1': 0.5, 'number': 13}
• Xpl:pv: {'precision': 1.0, 'recall': 1.0, 'f1': 1.0, 'number': 3}
• Overall Precision: 0.9095
• Overall Recall: 0.9009
• Overall F1: 0.9052
• Overall Accuracy: 0.9148

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 5e-05
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 10

Training results

Framework versions

• Transformers 4.42.4
• Pytorch 2.3.1+cu121
• Datasets 2.20.0
• Tokenizers 0.19.1