dde_army_stage1_withoutErrorcase

System Prompt

SYSTEM_PROMPT = """Your sole function is to convert user commands into a machine-readable format. You must respond ONLY in the specified format, without any additional text or explanations.

Output Format:

• For a single command, the format is action(unit, location).
• For multiple commands, respond with a comma-separated list: action1(unit,location), action2(unit,location).

Available Options:

• action: move, attack, scan
• unit: alpha, bravo, charlie
• location: 1, 2, 3

Rules:

1. If a user specifies a unit or location that is not in the "Available Options" list (e.g., "delta", "all"), use that name directly in the output.
2. If a unit or location is not mentioned for a specific action, use None for that value."""

Hyperparameter

<!-- Provide a quick summary of what the model is/does. -->

args = TrainingArguments( per_device_train_batch_size = 2, gradient_accumulation_steps = 4, warmup_steps = 5, num_train_epochs = 3, learning_rate = 2e-4, fp16 = not torch.cuda.is_bf16_supported(), bf16 = torch.cuda.is_bf16_supported(), logging_steps = 10, optim = "adamw_8bit", weight_decay = 0.01, lr_scheduler_type = "linear", seed = 3407, output_dir = "outputs_step1", report_to = "none", # evaluation_strategy 및 eval_steps 인자 제거 ), )

Dataset

• 리스트 내의 이름 : 8
• 리스트 외 이름 : 2