## Training hyperparameters
                               ```yaml
                               learning_rate: 5e-05
                               num_training_examples: 14000
                               batch_size_per_device: 1
                               gradient_accumulation_steps: 8
                               effective_batch_size: 8
                               max_steps: 1314
                               num_epochs: 3
                               optimizer: adamw
                               num_warmup_steps: 39
                               lr_scheduler_type: linear
                               max_seq_length: 1060
                               precision: bf16
                               gradient_checkpointing: true
                               seed:42
                               ```