{
  "num_iterations": 1,
  "num_steps": 312,
  "batch_size": 8,
  "num_generations": 8,
  "max_completion_length": 400,
  "beta": 0.005,
  "learning_rate": 5e-06,
  "mu": 1,
  "epsilon": 0.1,
  "lambda_blend": 0.1,
  "lambda_min": 0.01,
  "lambda_max": 0.3,
  "tau_H": 1.0,
  "entropy_ema_decay": 0.95,
  "temperature_m3po": 0.1,
  "use_m3po": true,
  "gating_type": "luong",
  "gating_config": {
    "temperature": 0.1,
    "rank": 256,
    "attn_dim": 256,
    "init_strategy": "xavier",
    "debug": false
  },
  "gating_warmup_steps": 50,
  "gating_lr": 5e-06,
  "gating_grad_clip": 1.0,
  "gradient_accumulation_steps": 2,
  "warmup_ratio": 0.1,
  "seed": 123,
  "trial_number": 1,
  "model_name": "Qwen/Qwen2.5-1.5B-Instruct"
}