ST01.json

{
    "alias": "vumblebot",
    "model": {
        "model_name_or_path": "monologg/koelectra-base-v3-finetuned-korquad",
        "retriever_name": "ATIREBM25",
        "reader_name": "CNN",
        "config_name": "",
        "tokenizer_name": ""
    },
    "data": {
        "dataset_name": "squad_kor_v1",
        "sub_datasets": "",
        "sub_datasets_ratio": "",
        "overwrite_cache": false,
        "preprocessing_num_workers": 2,
        "max_seq_length": 384,
        "pad_to_max_length": false,
        "doc_stride": 128,
        "max_answer_length": 30
    },
    "train": {
        "masking_ratio": 0.0,
        "do_train": true,
        "do_eval": true,
        "do_eval_during_training": true,
        "eval_step": 100,
        "pororo_prediction": false,
        "save_total_limit": 1,
        "save_steps": 100,
        "logging_steps": 100,
        "overwrite_output_dir": true,
        "freeze_backbone": false,
        "report_to": ["wandb"]
    },
    "retriever": {
        "b": 0.01,
        "k1": 0.1,
        "topk": 3,
        "alpha": 0.1,
        "retrain": false,
        "weight_decay": 0.01,
        "learning_rate": 3e-5,
        "num_train_epochs": 2,
        "per_device_eval_batch_size": 2,
        "gradient_accumulation_steps": 1,
        "per_device_train_batch_size": 4,
        "dense_train_dataset": "squad_kor_v1"
    }
}