training_args.json

{
  "model_name": "Qwen/Qwen3-32B",
  "max_seq_length": 4096,
  "load_in_4bit": true,
  "load_in_8bit": false,
  "lora_r": 32,
  "lora_alpha": 32,
  "lora_dropout": 0.05,
  "token": null,
  "dataset_name": "b-mc2/sql-create-context",
  "max_samples": 500,
  "learning_rate": 2e-05,
  "warmup_ratio": 0.1,
  "weight_decay": 0.1,
  "batch_size": 1,
  "gradient_accumulation_steps": 4,
  "num_generations": 2,
  "max_steps": 500,
  "output_dir": "continuous_awareness_outputs",
  "save_steps": 50,
  "logging_steps": 10,
  "seed": 3407,
  "save_peft": true,
  "save_merged": true,
  "reward_weights": {
    "metacognitive": 0.2,
    "alignment": 0.25,
    "grounding": 0.2,
    "interface": 0.15,
    "continuous": 0.2
  }
}