{
  "run": "pilot_max24_no_curriculum",
  "arm": "max24_no_curriculum",
  "args": {
    "run_name": "pilot_max24_no_curriculum",
    "arm": "max24_no_curriculum",
    "model_id": "Qwen/Qwen3-4B",
    "seed": 123,
    "modulus": 97,
    "stage_max_steps": "24",
    "stage_min_lengths": "1",
    "stage_train_max_lengths": "24",
    "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": 24,
      "steps": 75
    }
  ],
  "train_log": [
    {
      "run": "pilot_max24_no_curriculum",
      "arm": "max24_no_curriculum",
      "stage": "stage1_max24",
      "stage_idx": 1,
      "local_step": 1,
      "global_step": 1,
      "compiler_max_steps": 24,
      "elapsed_sec": 2.587,
      "expansion_event": "init",
      "init_loss": 4.590240001678467,
      "op_loss": 1.1039576530456543,
      "arg_loss": 4.577983379364014,
      "state_loss": 4.57909631729126,
      "state_train_accuracy": 0.0,
      "executor_loss": 4.579453945159912,
      "loss": 46.935401916503906
    },
    {
      "run": "pilot_max24_no_curriculum",
      "arm": "max24_no_curriculum",
      "stage": "stage1_max24",
      "stage_idx": 1,
      "local_step": 15,
      "global_step": 15,
      "compiler_max_steps": 24,
      "elapsed_sec": 24.466,
      "expansion_event": "init",
      "init_loss": 4.47072172164917,
      "op_loss": 1.094483494758606,
      "arg_loss": 4.057218074798584,
      "state_loss": 4.576575756072998,
      "state_train_accuracy": 0.0,
      "executor_loss": 4.57720422744751,
      "loss": 44.36002731323242
    },
    {
      "run": "pilot_max24_no_curriculum",
      "arm": "max24_no_curriculum",
      "stage": "stage1_max24",
      "stage_idx": 1,
      "local_step": 30,
      "global_step": 30,
      "compiler_max_steps": 24,
      "elapsed_sec": 48.472,
      "expansion_event": "init",
      "init_loss": 3.945693254470825,
      "op_loss": 1.0834176540374756,
      "arg_loss": 3.7396697998046875,
      "state_loss": 4.5763349533081055,
      "state_train_accuracy": 0.0,
      "executor_loss": 4.576767444610596,
      "loss": 40.97797393798828
    },
    {
      "run": "pilot_max24_no_curriculum",
      "arm": "max24_no_curriculum",
      "stage": "stage1_max24",
      "stage_idx": 1,
      "local_step": 45,
      "global_step": 45,
      "compiler_max_steps": 24,
      "elapsed_sec": 71.936,
      "expansion_event": "init",
      "init_loss": 4.773558139801025,
      "op_loss": 1.0905687808990479,
      "arg_loss": 3.59173583984375,
      "state_loss": 4.571274280548096,
      "state_train_accuracy": 0.0,
      "executor_loss": 4.565242290496826,
      "loss": 43.68825912475586
    },
    {
      "run": "pilot_max24_no_curriculum",
      "arm": "max24_no_curriculum",
      "stage": "stage1_max24",
      "stage_idx": 1,
      "local_step": 60,
      "global_step": 60,
      "compiler_max_steps": 24,
      "elapsed_sec": 95.423,
      "expansion_event": "init",
      "init_loss": 4.101237773895264,
      "op_loss": 1.1090598106384277,
      "arg_loss": 3.442962646484375,
      "state_loss": 4.565454483032227,
      "state_train_accuracy": 0.0,
      "executor_loss": 4.577531814575195,
      "loss": 40.42884826660156
    },
    {
      "run": "pilot_max24_no_curriculum",
      "arm": "max24_no_curriculum",
      "stage": "stage1_max24",
      "stage_idx": 1,
      "local_step": 75,
      "global_step": 75,
      "compiler_max_steps": 24,
      "elapsed_sec": 119.343,
      "expansion_event": "init",
      "init_loss": 3.6585161685943604,
      "op_loss": 1.1255985498428345,
      "arg_loss": 3.452875852584839,
      "state_loss": 4.574269771575928,
      "state_train_accuracy": 0.0,
      "executor_loss": 4.574923515319824,
      "loss": 38.720359802246094
    }
  ],
  "metrics": [
    {
      "stage": "stage1_max24",
      "split": "standard_L8",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.010307593271136284,
      "direct_accuracy": NaN,
      "init_accuracy": 0.25,
      "op_accuracy": 0.296875,
      "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,
      "run": "pilot_max24_no_curriculum",
      "arm": "max24_no_curriculum",
      "global_step": 75,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "paraphrase_L8",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.010307349264621735,
      "direct_accuracy": NaN,
      "init_accuracy": 0.0,
      "op_accuracy": 0.328125,
      "arg_accuracy": 0.03125,
      "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_max24_no_curriculum",
      "arm": "max24_no_curriculum",
      "global_step": 75,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "heldout_L8",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.010307177901268005,
      "direct_accuracy": NaN,
      "init_accuracy": 0.125,
      "op_accuracy": 0.34375,
      "arg_accuracy": 0.078125,
      "program_exact": 0.0,
      "state_accuracy": 0.03125,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.015625,
      "state_mean_correct_prefix": 0.125,
      "run": "pilot_max24_no_curriculum",
      "arm": "max24_no_curriculum",
      "global_step": 75,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "paired_L8",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.010308642871677876,
      "direct_accuracy": NaN,
      "init_accuracy": 0.0,
      "op_accuracy": 0.375,
      "arg_accuracy": 0.015625,
      "program_exact": 0.0,
      "state_accuracy": 0.015625,
      "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_max24_no_curriculum",
      "arm": "max24_no_curriculum",
      "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": 0.010306912008672953,
      "direct_accuracy": NaN,
      "init_accuracy": 0.125,
      "op_accuracy": 0.453125,
      "arg_accuracy": 0.0,
      "program_exact": 0.0,
      "state_accuracy": 0.015625,
      "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_max24_no_curriculum",
      "arm": "max24_no_curriculum",
      "global_step": 75,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "standard_L16",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.010307446587830782,
      "direct_accuracy": NaN,
      "init_accuracy": 0.0,
      "op_accuracy": 0.3125,
      "arg_accuracy": 0.0234375,
      "program_exact": 0.0,
      "state_accuracy": 0.0234375,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "run": "pilot_max24_no_curriculum",
      "arm": "max24_no_curriculum",
      "global_step": 75,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "paraphrase_L16",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.010308006312698126,
      "direct_accuracy": NaN,
      "init_accuracy": 0.125,
      "op_accuracy": 0.28125,
      "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,
      "run": "pilot_max24_no_curriculum",
      "arm": "max24_no_curriculum",
      "global_step": 75,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "heldout_L16",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.010307913646101952,
      "direct_accuracy": NaN,
      "init_accuracy": 0.0,
      "op_accuracy": 0.34375,
      "arg_accuracy": 0.046875,
      "program_exact": 0.0,
      "state_accuracy": 0.0078125,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "run": "pilot_max24_no_curriculum",
      "arm": "max24_no_curriculum",
      "global_step": 75,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "paired_L16",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.010307217482477427,
      "direct_accuracy": NaN,
      "init_accuracy": 0.0,
      "op_accuracy": 0.3125,
      "arg_accuracy": 0.0625,
      "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_max24_no_curriculum",
      "arm": "max24_no_curriculum",
      "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": 0.010309252422302961,
      "direct_accuracy": NaN,
      "init_accuracy": 0.0,
      "op_accuracy": 0.2265625,
      "arg_accuracy": 0.0390625,
      "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_max24_no_curriculum",
      "arm": "max24_no_curriculum",
      "global_step": 75,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "standard_L24",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.01030734134837985,
      "direct_accuracy": NaN,
      "init_accuracy": 0.0,
      "op_accuracy": 0.359375,
      "arg_accuracy": 0.057291666666666664,
      "program_exact": 0.0,
      "state_accuracy": 0.005208333333333333,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "run": "pilot_max24_no_curriculum",
      "arm": "max24_no_curriculum",
      "global_step": 75,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "paraphrase_L24",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.010307024698704481,
      "direct_accuracy": NaN,
      "init_accuracy": 0.0,
      "op_accuracy": 0.2604166666666667,
      "arg_accuracy": 0.020833333333333332,
      "program_exact": 0.0,
      "state_accuracy": 0.005208333333333333,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "run": "pilot_max24_no_curriculum",
      "arm": "max24_no_curriculum",
      "global_step": 75,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "heldout_L24",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.010307596996426582,
      "direct_accuracy": NaN,
      "init_accuracy": 0.125,
      "op_accuracy": 0.3541666666666667,
      "arg_accuracy": 0.026041666666666668,
      "program_exact": 0.0,
      "state_accuracy": 0.005208333333333333,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.0,
      "state_mean_correct_prefix": 0.0,
      "run": "pilot_max24_no_curriculum",
      "arm": "max24_no_curriculum",
      "global_step": 75,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "paired_L24",
      "n": 8,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.010307244956493378,
      "direct_accuracy": NaN,
      "init_accuracy": 0.0,
      "op_accuracy": 0.3333333333333333,
      "arg_accuracy": 0.07291666666666667,
      "program_exact": 0.0,
      "state_accuracy": 0.015625,
      "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_max24_no_curriculum",
      "arm": "max24_no_curriculum",
      "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": 0.010307100601494312,
      "direct_accuracy": NaN,
      "init_accuracy": 0.0,
      "op_accuracy": 0.2760416666666667,
      "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,
      "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_max24_no_curriculum",
      "arm": "max24_no_curriculum",
      "global_step": 75,
      "stage_idx": 1
    }
  ],
  "checkpoints": [],
  "elapsed_sec": 134.728
}