MishanyaP/dl2-hw2-ner

DL2 HW2 NER

BGE fine-tuned on the labeled training split. Hyperparameters selected on a validation subset. The test split was not used.

{ "loss": 0.07933095842599869, "precision": 0.9082387765170202, "recall": 0.9294850218781555, "f1": 0.9187390834234385, "accuracy": 0.9818259805868429, "runtime": 2.5699, "samples_per_second": 1264.646, "steps_per_second": 39.69, "name": "without-synthetic", "train_size": 10000, "seconds": 403.240630106, "params": { "learning_rate": 5.373713206635395e-05, "per_device_train_batch_size": 4 } }