{ | |
"experiment": "POS_Tagging", | |
"name": "ota_boun", | |
"n_gpu": 1, | |
"model": { | |
"type": "MultiParser", | |
"args": { | |
"embeddings_processor": { | |
"type": "BertWrapper", | |
"args": { | |
"model_path": "data/pretrained_embeddings/bert-base-turkish-cased-nonpretrained", | |
"fine_tune": true, | |
"hidden_dropout": 0.2, | |
"attn_dropout": 0.2, | |
"output_dropout": 0.5, | |
"scalar_mix_layer_dropout": 0.1, | |
"token_mask_prob": 0.15 | |
} | |
}, | |
"outputs": { | |
"upos": { | |
"type": "SequenceTagger", | |
"args": { | |
"hidden_size": 0, | |
"input_dropout": 0.2, | |
"vocab": { | |
"type": "BasicVocab", | |
"args": { | |
"vocab_filename": "data/corpora/ota_boun/vocab/basic.vocab" | |
} | |
} | |
} | |
} | |
} | |
} | |
}, | |
"data_loaders": { | |
"type": "BucketedCONLLLoader", | |
"args": { | |
"annotation_layers": { | |
"upos": { | |
"type": "TagSequence", | |
"source_column": 3, | |
"args": { | |
"ignore_root": true | |
} | |
} | |
}, | |
"batch_size": 32, | |
"bucket_size": 512, | |
"max_tokens_per_batch": 156800, | |
"num_workers": 2 | |
}, | |
"paths": { | |
"train": "data/corpora/ota_boun/ota_boun-ud-train.conllu", | |
"dev": "data/corpora/ota_boun/tr_boun-ud-test.conllu" | |
} | |
}, | |
"trainer": { | |
"min_epochs": 15, | |
"max_epochs": 300, | |
"early_stop": 15, | |
"save_dir": "data/saved_models/", | |
"save_period": 20, | |
"verbosity": 2, | |
"validation_criterion": { | |
"metrics": { | |
"upos": "fscore" | |
}, | |
"weighting": "multiplicative" | |
}, | |
"optimizer": { | |
"type": "AdamW", | |
"args": { | |
"lr": 4e-05, | |
"weight_decay": 0.0 | |
} | |
}, | |
"lr_scheduler": { | |
"type": "LambdaLR", | |
"args": { | |
"lr_lambda": "SqrtSchedule(400)" | |
} | |
}, | |
"loss": { | |
"type": "CrossEntropyLoss", | |
"args": { | |
"ignore_index": -1 | |
} | |
} | |
} | |
} |