TrainingArguments Explorer

⚙️ Build Your Configuration

📁 Output & Logging ▼

output_dir

logging_steps

report_to

🏋️ Training Parameters ▼

num_train_epochs

max_steps

per_device_train_batch_size

per_device_eval_batch_size

gradient_accumulation_steps

📈 Learning Rate ▼

learning_rate

2e-5

lr_scheduler_type

warmup_ratio

0.1

weight_decay

0.01

💾 Evaluation & Checkpoints ▼

evaluation_strategy

save_strategy

load_best_model_at_end

save_total_limit

🔄 Reproducibility ▼

seed

dataloader_num_workers

Quick Presets

Generated Config

⚠️ Configuration Warnings

🧠 Quick Check

If you have limited GPU memory, which parameter should you adjust first?

Increase num_train_epochs

Decrease batch_size and increase gradient_accumulation_steps

Set evaluation_strategy to "no"

Reducing batch size lowers memory usage, while gradient accumulation maintains effective batch size for stable training.