{
  "run": "main_max24_no_curriculum_seed123",
  "arm": "max24_no_curriculum",
  "seed": 123,
  "args": {
    "run_name": "main_max24_no_curriculum_seed123",
    "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": "750",
    "train_examples": 512,
    "train_template_mode": "mixed",
    "eval_lengths": "8,16,24",
    "eval_examples": 64,
    "paired_eval_pairs": 32,
    "batch_size": 8,
    "grad_accum": 1,
    "eval_batch_size": 8,
    "max_length": 2048,
    "head_width": 512,
    "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": 50,
    "save_checkpoints": true,
    "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": 750
    }
  ],
  "train_log": [
    {
      "run": "main_max24_no_curriculum_seed123",
      "arm": "max24_no_curriculum",
      "seed": 123,
      "stage": "stage1_max24",
      "stage_idx": 1,
      "local_step": 1,
      "global_step": 1,
      "compiler_max_steps": 24,
      "elapsed_sec": 6.343,
      "expansion_event": "init",
      "init_loss": 4.625215530395508,
      "op_loss": 1.1202036142349243,
      "arg_loss": 4.554243564605713,
      "state_loss": 4.580015182495117,
      "state_train_accuracy": 0.0,
      "executor_loss": 4.580340385437012,
      "loss": 46.99839782714844
    },
    {
      "run": "main_max24_no_curriculum_seed123",
      "arm": "max24_no_curriculum",
      "seed": 123,
      "stage": "stage1_max24",
      "stage_idx": 1,
      "local_step": 50,
      "global_step": 50,
      "compiler_max_steps": 24,
      "elapsed_sec": 173.698,
      "expansion_event": "init",
      "init_loss": 4.343316078186035,
      "op_loss": 1.1201934814453125,
      "arg_loss": 3.671297550201416,
      "state_loss": 4.5747389793396,
      "state_train_accuracy": 0.0,
      "executor_loss": 4.575059413909912,
      "loss": 42.32844543457031
    },
    {
      "run": "main_max24_no_curriculum_seed123",
      "arm": "max24_no_curriculum",
      "seed": 123,
      "stage": "stage1_max24",
      "stage_idx": 1,
      "local_step": 100,
      "global_step": 100,
      "compiler_max_steps": 24,
      "elapsed_sec": 346.052,
      "expansion_event": "init",
      "init_loss": 2.4111828804016113,
      "op_loss": 0.3148062229156494,
      "arg_loss": 1.714956283569336,
      "state_loss": 4.291322708129883,
      "state_train_accuracy": 0.1780821979045868,
      "executor_loss": 4.5542802810668945,
      "loss": 25.66496467590332
    },
    {
      "run": "main_max24_no_curriculum_seed123",
      "arm": "max24_no_curriculum",
      "seed": 123,
      "stage": "stage1_max24",
      "stage_idx": 1,
      "local_step": 150,
      "global_step": 150,
      "compiler_max_steps": 24,
      "elapsed_sec": 515.972,
      "expansion_event": "init",
      "init_loss": 0.7904766798019409,
      "op_loss": 0.20348726212978363,
      "arg_loss": 0.5742195844650269,
      "state_loss": 1.771788477897644,
      "state_train_accuracy": 0.8313252925872803,
      "executor_loss": 1.9611732959747314,
      "loss": 9.395234107971191
    },
    {
      "run": "main_max24_no_curriculum_seed123",
      "arm": "max24_no_curriculum",
      "seed": 123,
      "stage": "stage1_max24",
      "stage_idx": 1,
      "local_step": 200,
      "global_step": 200,
      "compiler_max_steps": 24,
      "elapsed_sec": 686.004,
      "expansion_event": "init",
      "init_loss": 0.15927912294864655,
      "op_loss": 0.000492714112624526,
      "arg_loss": 0.005612588953226805,
      "state_loss": 0.19532433152198792,
      "state_train_accuracy": 1.0,
      "executor_loss": 0.22623765468597412,
      "loss": 1.081621527671814
    },
    {
      "run": "main_max24_no_curriculum_seed123",
      "arm": "max24_no_curriculum",
      "seed": 123,
      "stage": "stage1_max24",
      "stage_idx": 1,
      "local_step": 250,
      "global_step": 250,
      "compiler_max_steps": 24,
      "elapsed_sec": 856.555,
      "expansion_event": "init",
      "init_loss": 0.02333131432533264,
      "op_loss": 2.3715203496976756e-05,
      "arg_loss": 0.0009712076862342656,
      "state_loss": 0.029928600415587425,
      "state_train_accuracy": 1.0,
      "executor_loss": 0.034395694732666016,
      "loss": 0.1615580916404724
    },
    {
      "run": "main_max24_no_curriculum_seed123",
      "arm": "max24_no_curriculum",
      "seed": 123,
      "stage": "stage1_max24",
      "stage_idx": 1,
      "local_step": 300,
      "global_step": 300,
      "compiler_max_steps": 24,
      "elapsed_sec": 1027.662,
      "expansion_event": "init",
      "init_loss": 0.004485909827053547,
      "op_loss": 2.973030177599867e-06,
      "arg_loss": 0.0001556042698211968,
      "state_loss": 0.007165959570556879,
      "state_train_accuracy": 0.9999999403953552,
      "executor_loss": 0.00613110326230526,
      "loss": 0.03186609223484993
    },
    {
      "run": "main_max24_no_curriculum_seed123",
      "arm": "max24_no_curriculum",
      "seed": 123,
      "stage": "stage1_max24",
      "stage_idx": 1,
      "local_step": 350,
      "global_step": 350,
      "compiler_max_steps": 24,
      "elapsed_sec": 1197.7,
      "expansion_event": "init",
      "init_loss": 0.0019030070398002863,
      "op_loss": 3.3880519367812667e-07,
      "arg_loss": 0.0001391927944496274,
      "state_loss": 0.0023219739086925983,
      "state_train_accuracy": 1.0,
      "executor_loss": 0.0032285992056131363,
      "loss": 0.01371971145272255
    },
    {
      "run": "main_max24_no_curriculum_seed123",
      "arm": "max24_no_curriculum",
      "seed": 123,
      "stage": "stage1_max24",
      "stage_idx": 1,
      "local_step": 400,
      "global_step": 400,
      "compiler_max_steps": 24,
      "elapsed_sec": 1368.773,
      "expansion_event": "init",
      "init_loss": 0.0007173412013798952,
      "op_loss": 0.1385205239057541,
      "arg_loss": 0.02241048403084278,
      "state_loss": 0.15111802518367767,
      "state_train_accuracy": 0.9789474010467529,
      "executor_loss": 0.38322651386260986,
      "loss": 0.7653763890266418
    },
    {
      "run": "main_max24_no_curriculum_seed123",
      "arm": "max24_no_curriculum",
      "seed": 123,
      "stage": "stage1_max24",
      "stage_idx": 1,
      "local_step": 450,
      "global_step": 450,
      "compiler_max_steps": 24,
      "elapsed_sec": 1539.088,
      "expansion_event": "init",
      "init_loss": 0.00037986435927450657,
      "op_loss": 1.355146309833799e-06,
      "arg_loss": 3.953124542022124e-05,
      "state_loss": 0.0005959459231235087,
      "state_train_accuracy": 1.0,
      "executor_loss": 0.0008246286306530237,
      "loss": 0.0030995120760053396
    },
    {
      "run": "main_max24_no_curriculum_seed123",
      "arm": "max24_no_curriculum",
      "seed": 123,
      "stage": "stage1_max24",
      "stage_idx": 1,
      "local_step": 500,
      "global_step": 500,
      "compiler_max_steps": 24,
      "elapsed_sec": 1710.029,
      "expansion_event": "init",
      "init_loss": 0.0003054669650737196,
      "op_loss": 5.489896466315258e-07,
      "arg_loss": 1.6718397091608495e-05,
      "state_loss": 0.00043837560224346817,
      "state_train_accuracy": 1.0,
      "executor_loss": 0.0005516228848136961,
      "loss": 0.0022792890667915344
    },
    {
      "run": "main_max24_no_curriculum_seed123",
      "arm": "max24_no_curriculum",
      "seed": 123,
      "stage": "stage1_max24",
      "stage_idx": 1,
      "local_step": 550,
      "global_step": 550,
      "compiler_max_steps": 24,
      "elapsed_sec": 1880.321,
      "expansion_event": "init",
      "init_loss": 0.00023348465037997812,
      "op_loss": 3.3161286410177127e-06,
      "arg_loss": 0.0002048293245024979,
      "state_loss": 0.002477017231285572,
      "state_train_accuracy": 0.9999999403953552,
      "executor_loss": 0.001664513722062111,
      "loss": 0.00589810311794281
    },
    {
      "run": "main_max24_no_curriculum_seed123",
      "arm": "max24_no_curriculum",
      "seed": 123,
      "stage": "stage1_max24",
      "stage_idx": 1,
      "local_step": 600,
      "global_step": 600,
      "compiler_max_steps": 24,
      "elapsed_sec": 2050.222,
      "expansion_event": "init",
      "init_loss": 0.0009104575146920979,
      "op_loss": 0.27364975214004517,
      "arg_loss": 0.614953875541687,
      "state_loss": 0.8363456726074219,
      "state_train_accuracy": 0.8255813717842102,
      "executor_loss": 1.8560597896575928,
      "loss": 5.4295125007629395
    },
    {
      "run": "main_max24_no_curriculum_seed123",
      "arm": "max24_no_curriculum",
      "seed": 123,
      "stage": "stage1_max24",
      "stage_idx": 1,
      "local_step": 650,
      "global_step": 650,
      "compiler_max_steps": 24,
      "elapsed_sec": 2222.289,
      "expansion_event": "init",
      "init_loss": 0.0009432347724214196,
      "op_loss": 0.14848676323890686,
      "arg_loss": 0.3924056887626648,
      "state_loss": 0.7844796180725098,
      "state_train_accuracy": 0.8360655307769775,
      "executor_loss": 2.9594645500183105,
      "loss": 5.465826511383057
    },
    {
      "run": "main_max24_no_curriculum_seed123",
      "arm": "max24_no_curriculum",
      "seed": 123,
      "stage": "stage1_max24",
      "stage_idx": 1,
      "local_step": 700,
      "global_step": 700,
      "compiler_max_steps": 24,
      "elapsed_sec": 2393.273,
      "expansion_event": "init",
      "init_loss": 0.0004934421740472317,
      "op_loss": 8.606372284702957e-05,
      "arg_loss": 0.0008870699675753713,
      "state_loss": 0.004874186124652624,
      "state_train_accuracy": 1.0,
      "executor_loss": 0.014592329040169716,
      "loss": 0.025074629113078117
    },
    {
      "run": "main_max24_no_curriculum_seed123",
      "arm": "max24_no_curriculum",
      "seed": 123,
      "stage": "stage1_max24",
      "stage_idx": 1,
      "local_step": 750,
      "global_step": 750,
      "compiler_max_steps": 24,
      "elapsed_sec": 2563.47,
      "expansion_event": "init",
      "init_loss": 0.0004732864326797426,
      "op_loss": 7.501017080357997e-06,
      "arg_loss": 0.0003299853706266731,
      "state_loss": 0.0024973817635327578,
      "state_train_accuracy": 1.0,
      "executor_loss": 0.004902468528598547,
      "loss": 0.010620438493788242
    }
  ],
  "metrics": [
    {
      "stage": "stage1_max24",
      "split": "standard_L8",
      "n": 64,
      "max_steps": 24,
      "executor_accuracy": 1.0,
      "executor_target_mass": 0.998943641781807,
      "direct_accuracy": NaN,
      "init_accuracy": 1.0,
      "op_accuracy": 1.0,
      "arg_accuracy": 1.0,
      "program_exact": 1.0,
      "state_accuracy": 1.0,
      "state_all_exact": 1.0,
      "state_prefix_fraction": 1.0,
      "state_mean_correct_prefix": 8.0,
      "run": "main_max24_no_curriculum_seed123",
      "arm": "max24_no_curriculum",
      "seed": 123,
      "global_step": 750,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "paraphrase_L8",
      "n": 64,
      "max_steps": 24,
      "executor_accuracy": 1.0,
      "executor_target_mass": 0.9989941120147705,
      "direct_accuracy": NaN,
      "init_accuracy": 1.0,
      "op_accuracy": 1.0,
      "arg_accuracy": 1.0,
      "program_exact": 1.0,
      "state_accuracy": 1.0,
      "state_all_exact": 1.0,
      "state_prefix_fraction": 1.0,
      "state_mean_correct_prefix": 8.0,
      "run": "main_max24_no_curriculum_seed123",
      "arm": "max24_no_curriculum",
      "seed": 123,
      "global_step": 750,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "heldout_L8",
      "n": 64,
      "max_steps": 24,
      "executor_accuracy": 0.921875,
      "executor_target_mass": 0.9073586314916611,
      "direct_accuracy": NaN,
      "init_accuracy": 1.0,
      "op_accuracy": 0.970703125,
      "arg_accuracy": 0.9609375,
      "program_exact": 0.921875,
      "state_accuracy": 0.955078125,
      "state_all_exact": 0.921875,
      "state_prefix_fraction": 0.955078125,
      "state_mean_correct_prefix": 7.640625,
      "run": "main_max24_no_curriculum_seed123",
      "arm": "max24_no_curriculum",
      "seed": 123,
      "global_step": 750,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "paired_L8",
      "n": 64,
      "max_steps": 24,
      "executor_accuracy": 1.0,
      "executor_target_mass": 0.9974109977483749,
      "direct_accuracy": NaN,
      "init_accuracy": 1.0,
      "op_accuracy": 1.0,
      "arg_accuracy": 1.0,
      "program_exact": 1.0,
      "state_accuracy": 1.0,
      "state_all_exact": 1.0,
      "state_prefix_fraction": 1.0,
      "state_mean_correct_prefix": 8.0,
      "executor_pair_answer_consistency": 1.0,
      "executor_pair_both_correct": 1.0,
      "compiler_pair_program_consistency": 1.0,
      "compiler_pair_state_consistency": 1.0,
      "run": "main_max24_no_curriculum_seed123",
      "arm": "max24_no_curriculum",
      "seed": 123,
      "global_step": 750,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "paired_heldout_L8",
      "n": 64,
      "max_steps": 24,
      "executor_accuracy": 0.96875,
      "executor_target_mass": 0.9669450893998146,
      "direct_accuracy": NaN,
      "init_accuracy": 1.0,
      "op_accuracy": 0.98828125,
      "arg_accuracy": 0.984375,
      "program_exact": 0.96875,
      "state_accuracy": 0.978515625,
      "state_all_exact": 0.96875,
      "state_prefix_fraction": 0.978515625,
      "state_mean_correct_prefix": 7.828125,
      "executor_pair_answer_consistency": 0.9375,
      "executor_pair_both_correct": 0.9375,
      "compiler_pair_program_consistency": 0.9375,
      "compiler_pair_state_consistency": 0.9375,
      "run": "main_max24_no_curriculum_seed123",
      "arm": "max24_no_curriculum",
      "seed": 123,
      "global_step": 750,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "standard_L16",
      "n": 64,
      "max_steps": 24,
      "executor_accuracy": 0.484375,
      "executor_target_mass": 0.4759833812713623,
      "direct_accuracy": NaN,
      "init_accuracy": 1.0,
      "op_accuracy": 0.9658203125,
      "arg_accuracy": 0.986328125,
      "program_exact": 0.453125,
      "state_accuracy": 0.9033203125,
      "state_all_exact": 0.484375,
      "state_prefix_fraction": 0.9033203125,
      "state_mean_correct_prefix": 14.453125,
      "run": "main_max24_no_curriculum_seed123",
      "arm": "max24_no_curriculum",
      "seed": 123,
      "global_step": 750,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "paraphrase_L16",
      "n": 64,
      "max_steps": 24,
      "executor_accuracy": 0.953125,
      "executor_target_mass": 0.9558496549725533,
      "direct_accuracy": NaN,
      "init_accuracy": 1.0,
      "op_accuracy": 0.998046875,
      "arg_accuracy": 0.9990234375,
      "program_exact": 0.953125,
      "state_accuracy": 0.9892578125,
      "state_all_exact": 0.953125,
      "state_prefix_fraction": 0.9892578125,
      "state_mean_correct_prefix": 15.828125,
      "run": "main_max24_no_curriculum_seed123",
      "arm": "max24_no_curriculum",
      "seed": 123,
      "global_step": 750,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "heldout_L16",
      "n": 64,
      "max_steps": 24,
      "executor_accuracy": 0.703125,
      "executor_target_mass": 0.6821616776287556,
      "direct_accuracy": NaN,
      "init_accuracy": 1.0,
      "op_accuracy": 0.962890625,
      "arg_accuracy": 0.9521484375,
      "program_exact": 0.6875,
      "state_accuracy": 0.9375,
      "state_all_exact": 0.6875,
      "state_prefix_fraction": 0.935546875,
      "state_mean_correct_prefix": 14.96875,
      "run": "main_max24_no_curriculum_seed123",
      "arm": "max24_no_curriculum",
      "seed": 123,
      "global_step": 750,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "paired_L16",
      "n": 64,
      "max_steps": 24,
      "executor_accuracy": 0.671875,
      "executor_target_mass": 0.6704260632395744,
      "direct_accuracy": NaN,
      "init_accuracy": 1.0,
      "op_accuracy": 0.98046875,
      "arg_accuracy": 0.9892578125,
      "program_exact": 0.671875,
      "state_accuracy": 0.93359375,
      "state_all_exact": 0.671875,
      "state_prefix_fraction": 0.93359375,
      "state_mean_correct_prefix": 14.9375,
      "executor_pair_answer_consistency": 0.375,
      "executor_pair_both_correct": 0.375,
      "compiler_pair_program_consistency": 0.375,
      "compiler_pair_state_consistency": 0.375,
      "run": "main_max24_no_curriculum_seed123",
      "arm": "max24_no_curriculum",
      "seed": 123,
      "global_step": 750,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "paired_heldout_L16",
      "n": 64,
      "max_steps": 24,
      "executor_accuracy": 0.6875,
      "executor_target_mass": 0.677246943116188,
      "direct_accuracy": NaN,
      "init_accuracy": 1.0,
      "op_accuracy": 0.9716796875,
      "arg_accuracy": 0.9736328125,
      "program_exact": 0.6875,
      "state_accuracy": 0.9169921875,
      "state_all_exact": 0.6875,
      "state_prefix_fraction": 0.9169921875,
      "state_mean_correct_prefix": 14.671875,
      "executor_pair_answer_consistency": 0.40625,
      "executor_pair_both_correct": 0.40625,
      "compiler_pair_program_consistency": 0.40625,
      "compiler_pair_state_consistency": 0.40625,
      "run": "main_max24_no_curriculum_seed123",
      "arm": "max24_no_curriculum",
      "seed": 123,
      "global_step": 750,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "standard_L24",
      "n": 64,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 0.01106625929242,
      "direct_accuracy": NaN,
      "init_accuracy": 1.0,
      "op_accuracy": 0.8287760416666666,
      "arg_accuracy": 0.7552083333333334,
      "program_exact": 0.0,
      "state_accuracy": 0.623046875,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.6217447916666666,
      "state_mean_correct_prefix": 14.921875,
      "run": "main_max24_no_curriculum_seed123",
      "arm": "max24_no_curriculum",
      "seed": 123,
      "global_step": 750,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "paraphrase_L24",
      "n": 64,
      "max_steps": 24,
      "executor_accuracy": 0.46875,
      "executor_target_mass": 0.4189531244337559,
      "direct_accuracy": NaN,
      "init_accuracy": 1.0,
      "op_accuracy": 0.982421875,
      "arg_accuracy": 0.9700520833333334,
      "program_exact": 0.46875,
      "state_accuracy": 0.7877604166666666,
      "state_all_exact": 0.46875,
      "state_prefix_fraction": 0.7877604166666666,
      "state_mean_correct_prefix": 18.90625,
      "run": "main_max24_no_curriculum_seed123",
      "arm": "max24_no_curriculum",
      "seed": 123,
      "global_step": 750,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "heldout_L24",
      "n": 64,
      "max_steps": 24,
      "executor_accuracy": 0.109375,
      "executor_target_mass": 0.07105528621468693,
      "direct_accuracy": NaN,
      "init_accuracy": 1.0,
      "op_accuracy": 0.8684895833333334,
      "arg_accuracy": 0.8151041666666666,
      "program_exact": 0.109375,
      "state_accuracy": 0.6627604166666666,
      "state_all_exact": 0.109375,
      "state_prefix_fraction": 0.6614583333333334,
      "state_mean_correct_prefix": 15.875,
      "run": "main_max24_no_curriculum_seed123",
      "arm": "max24_no_curriculum",
      "seed": 123,
      "global_step": 750,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "paired_L24",
      "n": 64,
      "max_steps": 24,
      "executor_accuracy": 0.234375,
      "executor_target_mass": 0.19069256633520126,
      "direct_accuracy": NaN,
      "init_accuracy": 1.0,
      "op_accuracy": 0.8899739583333334,
      "arg_accuracy": 0.8509114583333334,
      "program_exact": 0.203125,
      "state_accuracy": 0.716796875,
      "state_all_exact": 0.203125,
      "state_prefix_fraction": 0.7135416666666666,
      "state_mean_correct_prefix": 17.125,
      "executor_pair_answer_consistency": 0.03125,
      "executor_pair_both_correct": 0.03125,
      "compiler_pair_program_consistency": 0.0,
      "compiler_pair_state_consistency": 0.0,
      "run": "main_max24_no_curriculum_seed123",
      "arm": "max24_no_curriculum",
      "seed": 123,
      "global_step": 750,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "paired_heldout_L24",
      "n": 64,
      "max_steps": 24,
      "executor_accuracy": 0.046875,
      "executor_target_mass": 0.03860202612122521,
      "direct_accuracy": NaN,
      "init_accuracy": 1.0,
      "op_accuracy": 0.8567708333333334,
      "arg_accuracy": 0.7916666666666666,
      "program_exact": 0.046875,
      "state_accuracy": 0.6770833333333334,
      "state_all_exact": 0.046875,
      "state_prefix_fraction": 0.671875,
      "state_mean_correct_prefix": 16.125,
      "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": "main_max24_no_curriculum_seed123",
      "arm": "max24_no_curriculum",
      "seed": 123,
      "global_step": 750,
      "stage_idx": 1
    }
  ],
  "checkpoints": [
    {
      "run": "main_max24_no_curriculum_seed123",
      "seed": 123,
      "tag": "stage1_max24_step00750",
      "stage": "stage1_max24",
      "global_step": 750,
      "compiler_max_steps": 24,
      "checkpoint_dir": "/workspace/large_artifacts/qwen_compiler_multiseed_reattribution/checkpoints/main_max24_no_curriculum_seed123/stage1_max24_step00750",
      "model_checkpoint": "/workspace/large_artifacts/qwen_compiler_multiseed_reattribution/checkpoints/main_max24_no_curriculum_seed123/stage1_max24_step00750/adapter",
      "heads_checkpoint": "/workspace/large_artifacts/qwen_compiler_multiseed_reattribution/checkpoints/main_max24_no_curriculum_seed123/stage1_max24_step00750/heads.pt",
      "selection_split": "paired_L24",
      "selection_metric": "executor_pair_both_correct",
      "selection_value": 0.03125,
      "executor_accuracy": 0.234375,
      "program_exact": 0.203125,
      "state_prefix_fraction": 0.7135416666666666
    }
  ],
  "elapsed_sec": 2690.699
}