models_config.json

{
    "mon": {
        "dataset_path": "data/baby_cry_detection",
        "output_dir": "distilhubert-finetuned-cry-detector",
        "training_args": {
            "num_train_epochs": 10, 
            "learning_rate": 0.00003,
            "warmup_ratio": 0.1,
            "output_dir": "distilhubert-finetuned-cry-detector",
            "eval_strategy": "steps",
            "eval_steps": 100,
            "save_strategy": "steps",
            "save_steps": 100,
            "save_total_limit": 5,
            "lr_scheduler_type": "cosine_with_restarts",
            "auto_find_batch_size": true,
            "per_device_train_batch_size": 64,
            "per_device_eval_batch_size": 64,
            "gradient_accumulation_steps": 2,
            "gradient_checkpointing": true,
            "load_best_model_at_end": true,
            "greater_is_better": true,
            "metric_for_best_model": "eval_f1",
            "optim": "adamw_torch",
            "hub_strategy": "checkpoint",
            "report_to": "tensorboard",
            "full_determinism": true,
            "seed": 123,
            "data_seed":123,
            "max_grad_norm": 0.5,
            "weight_decay": 0.02,
            "fp16": true,
            "logging_steps": 50,
            "eval_accumulation_steps": 1,
            "dataloader_num_workers": 12,
            "dataloader_pin_memory": true,
            "label_smoothing_factor": 0.1,
            "length_column_name": "input_length"
        }
    },
    "class": {
        "dataset_path": "data/mixed_data",
        "output_dir": "distilhubert-finetuned-mixed-data",
        "training_args": {
            "num_train_epochs": 30,
            "learning_rate": 3e-4,
            "warmup_ratio": 0.1,
            "output_dir": "distilhubert-finetuned-mixed-data",
            "eval_strategy": "steps",
            "eval_steps": 100,
            "save_strategy": "steps",
            "save_steps": 100,
            "save_total_limit": 5,
            "lr_scheduler_type": "cosine_with_restarts",
            "auto_find_batch_size": true,
            "per_device_train_batch_size": 64,
            "per_device_eval_batch_size": 64,
            "gradient_accumulation_steps": 2,
            "gradient_checkpointing": true,
            "load_best_model_at_end": true,
            "greater_is_better": true,
            "metric_for_best_model": "eval_f1",
            "optim": "adamw_torch",
            "seed": 123,
            "data_seed": 123,
            "max_grad_norm": 0.5,
            "weight_decay": 0.02,
            "fp16": true,
            "logging_steps": 50,
            "eval_accumulation_steps": 1,
            "dataloader_num_workers": 12,
            "dataloader_pin_memory": true,
            "label_smoothing_factor": 0.1,
            "length_column_name": "input_length",
            "hub_strategy": "checkpoint",
            "report_to": "tensorboard"
        }
    }
}