{
  "run": "pilot_train_to16_eval24",
  "arm": "train_to16_eval24",
  "args": {
    "run_name": "pilot_train_to16_eval24",
    "arm": "train_to16_eval24",
    "model_id": "Qwen/Qwen3-4B",
    "seed": 123,
    "modulus": 97,
    "stage_max_steps": "24",
    "stage_min_lengths": "1",
    "stage_train_max_lengths": "16",
    "stage_steps": "75",
    "train_examples": 96,
    "train_template_mode": "mixed",
    "eval_lengths": "8,16,24",
    "eval_examples": 8,
    "paired_eval_pairs": 4,
    "batch_size": 4,
    "grad_accum": 1,
    "eval_batch_size": 4,
    "max_length": 2048,
    "head_width": 256,
    "compiler_layers": 1,
    "compiler_heads": 4,
    "compiler_dropout": 0.05,
    "expansion_noise": 0.005,
    "expansion_mode": "copy_last",
    "trace_loss_weight": 1.0,
    "executor_loss_weight": 1.0,
    "state_loss_weight": 1.0,
    "init_trace_loss_weight": 4.0,
    "op_trace_loss_weight": 1.0,
    "arg_trace_loss_weight": 4.0,
    "direct_head_weight": 0.0,
    "lr": 0.0002,
    "weight_decay": 0.0,
    "optimizer": "paged_adamw_8bit",
    "max_grad_norm": 1.0,
    "torch_dtype": "bf16",
    "load_in_4bit": true,
    "device_map": "auto",
    "use_lora": true,
    "gradient_checkpointing": true,
    "lora_r": 8,
    "lora_alpha": 16,
    "lora_dropout": 0.05,
    "lora_target_modules": "all-linear",
    "log_interval": 15,
    "save_checkpoints": false,
    "checkpoint_dir": "",
    "selection_split": "paired_L24",
    "selection_metric": "executor_pair_both_correct"
  },
  "metadata": {
    "python": "3.12.3",
    "platform": "Linux-6.8.0-62-generic-x86_64-with-glibc2.39",
    "torch": "2.8.0+cu128",
    "cuda_available": true,
    "transformers_loader": "AutoModelForCausalLM",
    "peft_installed": true,
    "use_lora": true,
    "lora_r": 8,
    "model_id": "Qwen/Qwen3-4B",
    "gpu_name": "NVIDIA RTX 6000 Ada Generation",
    "gpu_vram_gb": 47.383
  },
  "stages": [
    {
      "name": "stage1_max24",
      "max_steps": 24,
      "train_min_len": 1,
      "train_max_len": 16,
      "steps": 75
    }
  ],
  "train_log": [
    {
      "run": "pilot_train_to16_eval24",
      "arm": "train_to16_eval24",
      "stage": "stage1_max24",
      "stage_idx": 1,
      "local_step": 1,
      "global_step": 1,
      "compiler_max_steps": 24,
      "elapsed_sec": 2.375,
      "expansion_event": "init",
      "init_loss": 4.569614887237549,
      "op_loss": 1.0995908975601196,
      "arg_loss": 4.6720051765441895,
      "state_loss": 4.578830242156982,
      "state_train_accuracy": 0.0,
      "executor_loss": 4.5797200202941895,
      "loss": 47.2246208190918
    },
    {
      "run": "pilot_train_to16_eval24",
      "arm": "train_to16_eval24",
      "stage": "stage1_max24",
      "stage_idx": 1,
      "local_step": 15,
      "global_step": 15,
      "compiler_max_steps": 24,
      "elapsed_sec": 22.647,
      "expansion_event": "init",
      "init_loss": 4.295513153076172,
      "op_loss": 1.1061465740203857,
      "arg_loss": 4.065327167510986,
      "state_loss": 4.558757305145264,
      "state_train_accuracy": 0.05999999865889549,
      "executor_loss": 4.497153282165527,
      "loss": 43.60541915893555
    },
    {
      "run": "pilot_train_to16_eval24",
      "arm": "train_to16_eval24",
      "stage": "stage1_max24",
      "stage_idx": 1,
      "local_step": 30,
      "global_step": 30,
      "compiler_max_steps": 24,
      "elapsed_sec": 44.584,
      "expansion_event": "init",
      "init_loss": 4.496400356292725,
      "op_loss": 1.0982639789581299,
      "arg_loss": 3.8027150630950928,
      "state_loss": 4.577423095703125,
      "state_train_accuracy": 0.0,
      "executor_loss": 4.576974391937256,
      "loss": 43.44912338256836
    },
    {
      "run": "pilot_train_to16_eval24",
      "arm": "train_to16_eval24",
      "stage": "stage1_max24",
      "stage_idx": 1,
      "local_step": 45,
      "global_step": 45,
      "compiler_max_steps": 24,
      "elapsed_sec": 66.014,
      "expansion_event": "init",
      "init_loss": 4.757564067840576,
      "op_loss": 1.1051839590072632,
      "arg_loss": 3.7706477642059326,
      "state_loss": 4.570503234863281,
      "state_train_accuracy": 0.0,
      "executor_loss": 4.576455116271973,
      "loss": 44.364990234375
    },
    {
      "run": "pilot_train_to16_eval24",
      "arm": "train_to16_eval24",
      "stage": "stage1_max24",
      "stage_idx": 1,
      "local_step": 60,
      "global_step": 60,
      "compiler_max_steps": 24,
      "elapsed_sec": 87.959,
      "expansion_event": "init",
      "init_loss": 3.9635367393493652,
      "op_loss": 1.16115403175354,
      "arg_loss": 3.739800214767456,
      "state_loss": 4.577351093292236,
      "state_train_accuracy": 0.0,
      "executor_loss": 4.57549524307251,
      "loss": 41.127349853515625
    },
    {
      "run": "pilot_train_to16_eval24",
      "arm": "train_to16_eval24",
      "stage": "stage1_max24",
      "stage_idx": 1,
      "local_step": 75,
      "global_step": 75,
      "compiler_max_steps": 24,
      "elapsed_sec": 109.978,
      "expansion_event": "init",
      "init_loss": 4.062205791473389,
      "op_loss": 1.1204496622085571,
      "arg_loss": 3.625684976577759,
      "state_loss": 4.57267951965332,
      "state_train_accuracy": 0.02083333395421505,
      "executor_loss": 4.575016975402832,
      "loss": 41.01970672607422
    }
  ],
  "metrics": [
    {
      "stage": "stage1_max24",
      "split": "standard_L8",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.010305116884410381,
      "direct_accuracy": NaN,
      "init_accuracy": 0.125,
      "op_accuracy": 0.265625,
      "arg_accuracy": 0.03125,
      "program_exact": 0.0,
      "state_accuracy": 0.015625,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "run": "pilot_train_to16_eval24",
      "arm": "train_to16_eval24",
      "global_step": 75,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "paraphrase_L8",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.01030401373282075,
      "direct_accuracy": NaN,
      "init_accuracy": 0.0,
      "op_accuracy": 0.3125,
      "arg_accuracy": 0.0625,
      "program_exact": 0.0,
      "state_accuracy": 0.015625,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "run": "pilot_train_to16_eval24",
      "arm": "train_to16_eval24",
      "global_step": 75,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "heldout_L8",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.010304049123078585,
      "direct_accuracy": NaN,
      "init_accuracy": 0.375,
      "op_accuracy": 0.296875,
      "arg_accuracy": 0.046875,
      "program_exact": 0.0,
      "state_accuracy": 0.0,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "run": "pilot_train_to16_eval24",
      "arm": "train_to16_eval24",
      "global_step": 75,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "paired_L8",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.01030694181099534,
      "direct_accuracy": NaN,
      "init_accuracy": 0.25,
      "op_accuracy": 0.28125,
      "arg_accuracy": 0.0,
      "program_exact": 0.0,
      "state_accuracy": 0.0,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "executor_pair_answer_consistency": 0.25,
      "executor_pair_both_correct": 0.0,
      "compiler_pair_program_consistency": 0.25,
      "compiler_pair_state_consistency": 0.25,
      "run": "pilot_train_to16_eval24",
      "arm": "train_to16_eval24",
      "global_step": 75,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "paired_heldout_L8",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.010303620249032974,
      "direct_accuracy": NaN,
      "init_accuracy": 0.5,
      "op_accuracy": 0.359375,
      "arg_accuracy": 0.0,
      "program_exact": 0.0,
      "state_accuracy": 0.0,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "executor_pair_answer_consistency": 0.25,
      "executor_pair_both_correct": 0.0,
      "compiler_pair_program_consistency": 0.25,
      "compiler_pair_state_consistency": 0.25,
      "run": "pilot_train_to16_eval24",
      "arm": "train_to16_eval24",
      "global_step": 75,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "standard_L16",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.010302577633410692,
      "direct_accuracy": NaN,
      "init_accuracy": 0.25,
      "op_accuracy": 0.3203125,
      "arg_accuracy": 0.0390625,
      "program_exact": 0.0,
      "state_accuracy": 0.0,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "run": "pilot_train_to16_eval24",
      "arm": "train_to16_eval24",
      "global_step": 75,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "paraphrase_L16",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.010305809788405895,
      "direct_accuracy": NaN,
      "init_accuracy": 0.25,
      "op_accuracy": 0.4453125,
      "arg_accuracy": 0.0234375,
      "program_exact": 0.0,
      "state_accuracy": 0.015625,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "run": "pilot_train_to16_eval24",
      "arm": "train_to16_eval24",
      "global_step": 75,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "heldout_L16",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.010304210241883993,
      "direct_accuracy": NaN,
      "init_accuracy": 0.0,
      "op_accuracy": 0.296875,
      "arg_accuracy": 0.0625,
      "program_exact": 0.0,
      "state_accuracy": 0.015625,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "run": "pilot_train_to16_eval24",
      "arm": "train_to16_eval24",
      "global_step": 75,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "paired_L16",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.010303682181984186,
      "direct_accuracy": NaN,
      "init_accuracy": 0.0,
      "op_accuracy": 0.34375,
      "arg_accuracy": 0.0546875,
      "program_exact": 0.0,
      "state_accuracy": 0.0078125,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "executor_pair_answer_consistency": 0.0,
      "executor_pair_both_correct": 0.0,
      "compiler_pair_program_consistency": 0.0,
      "compiler_pair_state_consistency": 0.0,
      "run": "pilot_train_to16_eval24",
      "arm": "train_to16_eval24",
      "global_step": 75,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "paired_heldout_L16",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.01030876999720931,
      "direct_accuracy": NaN,
      "init_accuracy": 0.25,
      "op_accuracy": 0.2421875,
      "arg_accuracy": 0.0234375,
      "program_exact": 0.0,
      "state_accuracy": 0.0078125,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "executor_pair_answer_consistency": 0.0,
      "executor_pair_both_correct": 0.0,
      "compiler_pair_program_consistency": 0.0,
      "compiler_pair_state_consistency": 0.0,
      "run": "pilot_train_to16_eval24",
      "arm": "train_to16_eval24",
      "global_step": 75,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "standard_L24",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.010304977651685476,
      "direct_accuracy": NaN,
      "init_accuracy": 0.0,
      "op_accuracy": 0.3333333333333333,
      "arg_accuracy": 0.0625,
      "program_exact": 0.0,
      "state_accuracy": 0.010416666666666666,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "run": "pilot_train_to16_eval24",
      "arm": "train_to16_eval24",
      "global_step": 75,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "paraphrase_L24",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.010303618852049112,
      "direct_accuracy": NaN,
      "init_accuracy": 0.0,
      "op_accuracy": 0.359375,
      "arg_accuracy": 0.046875,
      "program_exact": 0.0,
      "state_accuracy": 0.010416666666666666,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "run": "pilot_train_to16_eval24",
      "arm": "train_to16_eval24",
      "global_step": 75,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "heldout_L24",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.010304729454219341,
      "direct_accuracy": NaN,
      "init_accuracy": 0.0,
      "op_accuracy": 0.3854166666666667,
      "arg_accuracy": 0.036458333333333336,
      "program_exact": 0.0,
      "state_accuracy": 0.010416666666666666,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "run": "pilot_train_to16_eval24",
      "arm": "train_to16_eval24",
      "global_step": 75,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "paired_L24",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.010304126888513565,
      "direct_accuracy": NaN,
      "init_accuracy": 0.0,
      "op_accuracy": 0.2708333333333333,
      "arg_accuracy": 0.0625,
      "program_exact": 0.0,
      "state_accuracy": 0.0,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "executor_pair_answer_consistency": 0.0,
      "executor_pair_both_correct": 0.0,
      "compiler_pair_program_consistency": 0.0,
      "compiler_pair_state_consistency": 0.0,
      "run": "pilot_train_to16_eval24",
      "arm": "train_to16_eval24",
      "global_step": 75,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "paired_heldout_L24",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.01030305027961731,
      "direct_accuracy": NaN,
      "init_accuracy": 0.0,
      "op_accuracy": 0.34375,
      "arg_accuracy": 0.041666666666666664,
      "program_exact": 0.0,
      "state_accuracy": 0.0,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "executor_pair_answer_consistency": 0.25,
      "executor_pair_both_correct": 0.0,
      "compiler_pair_program_consistency": 0.0,
      "compiler_pair_state_consistency": 0.0,
      "run": "pilot_train_to16_eval24",
      "arm": "train_to16_eval24",
      "global_step": 75,
      "stage_idx": 1
    }
  ],
  "checkpoints": [],
  "elapsed_sec": 125.261
}