azzenn4/

ModernBERT-6_Multiclass_Emotion

final_training_args = TrainingArguments(
    output_dir=outputs_dir,
    per_device_train_batch_size=32,
    num_train_epochs=3,
    weight_decay=0.01,
    learning_rate=3e-6,
    warmup_steps=700,
    lr_scheduler_type="cosine",
    eval_strategy="steps",
    eval_steps=100,
    save_steps=100,
    save_strategy="steps",
    logging_dir=logs_dir,
    logging_steps=100,
    load_best_model_at_end=True,
    metric_for_best_model="eval_loss",
    # report_to="none",
    fp16=True,
    disable_tqdm=False,
    max_grad_norm=10.2 
)