{ "batch_size": 32, "max_epochs": 10, "learning_rate": 2e-05, "warmup_proportion": 0.1, "gradient_accumulation_steps": 1, "weight_decay": 0.01, "early_stopping_patience": 3, "evaluation_steps": 100, "max_length": 128, "model_type": "mobilebert", "num_classes": 8, "num_workers": 4, "use_fp16": true, "output_dir": "output", "device": "cpu" }