{
  "run": "main_max24_no_curriculum_seed456",
  "arm": "max24_no_curriculum",
  "seed": 456,
  "args": {
    "run_name": "main_max24_no_curriculum_seed456",
    "arm": "max24_no_curriculum",
    "model_id": "Qwen/Qwen3-4B",
    "seed": 456,
    "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_seed456",
      "arm": "max24_no_curriculum",
      "seed": 456,
      "stage": "stage1_max24",
      "stage_idx": 1,
      "local_step": 1,
      "global_step": 1,
      "compiler_max_steps": 24,
      "elapsed_sec": 5.956,
      "expansion_event": "init",
      "init_loss": 4.546344757080078,
      "op_loss": 1.105911374092102,
      "arg_loss": 4.6274285316467285,
      "state_loss": 4.582851409912109,
      "state_train_accuracy": 0.0,
      "executor_loss": 4.5830278396606445,
      "loss": 46.966880798339844
    },
    {
      "run": "main_max24_no_curriculum_seed456",
      "arm": "max24_no_curriculum",
      "seed": 456,
      "stage": "stage1_max24",
      "stage_idx": 1,
      "local_step": 50,
      "global_step": 50,
      "compiler_max_steps": 24,
      "elapsed_sec": 172.683,
      "expansion_event": "init",
      "init_loss": 4.597607135772705,
      "op_loss": 1.0868052244186401,
      "arg_loss": 3.565945625305176,
      "state_loss": 4.574114799499512,
      "state_train_accuracy": 0.012195121496915817,
      "executor_loss": 4.575074672698975,
      "loss": 42.89020538330078
    },
    {
      "run": "main_max24_no_curriculum_seed456",
      "arm": "max24_no_curriculum",
      "seed": 456,
      "stage": "stage1_max24",
      "stage_idx": 1,
      "local_step": 100,
      "global_step": 100,
      "compiler_max_steps": 24,
      "elapsed_sec": 343.014,
      "expansion_event": "init",
      "init_loss": 4.399938583374023,
      "op_loss": 1.0965760946273804,
      "arg_loss": 3.6090705394744873,
      "state_loss": 4.574305057525635,
      "state_train_accuracy": 0.0,
      "executor_loss": 4.572355270385742,
      "loss": 42.27927780151367
    },
    {
      "run": "main_max24_no_curriculum_seed456",
      "arm": "max24_no_curriculum",
      "seed": 456,
      "stage": "stage1_max24",
      "stage_idx": 1,
      "local_step": 150,
      "global_step": 150,
      "compiler_max_steps": 24,
      "elapsed_sec": 513.451,
      "expansion_event": "init",
      "init_loss": 4.177130699157715,
      "op_loss": 1.09330415725708,
      "arg_loss": 3.6517157554626465,
      "state_loss": 4.574879169464111,
      "state_train_accuracy": 0.0,
      "executor_loss": 4.575400352478027,
      "loss": 41.5589714050293
    },
    {
      "run": "main_max24_no_curriculum_seed456",
      "arm": "max24_no_curriculum",
      "seed": 456,
      "stage": "stage1_max24",
      "stage_idx": 1,
      "local_step": 200,
      "global_step": 200,
      "compiler_max_steps": 24,
      "elapsed_sec": 684.79,
      "expansion_event": "init",
      "init_loss": 4.424065589904785,
      "op_loss": 1.109974980354309,
      "arg_loss": 3.589475393295288,
      "state_loss": 4.574131965637207,
      "state_train_accuracy": 0.016129031777381897,
      "executor_loss": 4.574737548828125,
      "loss": 42.31300735473633
    },
    {
      "run": "main_max24_no_curriculum_seed456",
      "arm": "max24_no_curriculum",
      "seed": 456,
      "stage": "stage1_max24",
      "stage_idx": 1,
      "local_step": 250,
      "global_step": 250,
      "compiler_max_steps": 24,
      "elapsed_sec": 856.097,
      "expansion_event": "init",
      "init_loss": 3.3248255252838135,
      "op_loss": 1.1070395708084106,
      "arg_loss": 3.540616989135742,
      "state_loss": 4.568655967712402,
      "state_train_accuracy": 0.008547008968889713,
      "executor_loss": 4.574756622314453,
      "loss": 37.712223052978516
    },
    {
      "run": "main_max24_no_curriculum_seed456",
      "arm": "max24_no_curriculum",
      "seed": 456,
      "stage": "stage1_max24",
      "stage_idx": 1,
      "local_step": 300,
      "global_step": 300,
      "compiler_max_steps": 24,
      "elapsed_sec": 1025.755,
      "expansion_event": "init",
      "init_loss": 1.2404069900512695,
      "op_loss": 1.091278076171875,
      "arg_loss": 3.4548280239105225,
      "state_loss": 4.54669189453125,
      "state_train_accuracy": 0.0,
      "executor_loss": 4.574949264526367,
      "loss": 28.993860244750977
    },
    {
      "run": "main_max24_no_curriculum_seed456",
      "arm": "max24_no_curriculum",
      "seed": 456,
      "stage": "stage1_max24",
      "stage_idx": 1,
      "local_step": 350,
      "global_step": 350,
      "compiler_max_steps": 24,
      "elapsed_sec": 1196.253,
      "expansion_event": "init",
      "init_loss": 0.089181087911129,
      "op_loss": 1.0962629318237305,
      "arg_loss": 2.800682544708252,
      "state_loss": 4.5189690589904785,
      "state_train_accuracy": 0.0078125,
      "executor_loss": 4.505682945251465,
      "loss": 21.680368423461914
    },
    {
      "run": "main_max24_no_curriculum_seed456",
      "arm": "max24_no_curriculum",
      "seed": 456,
      "stage": "stage1_max24",
      "stage_idx": 1,
      "local_step": 400,
      "global_step": 400,
      "compiler_max_steps": 24,
      "elapsed_sec": 1367.161,
      "expansion_event": "init",
      "init_loss": 0.056772224605083466,
      "op_loss": 2.691371992114e-05,
      "arg_loss": 0.007860989309847355,
      "state_loss": 0.10288829356431961,
      "state_train_accuracy": 0.9999999403953552,
      "executor_loss": 0.15221282839775085,
      "loss": 0.5136609077453613
    },
    {
      "run": "main_max24_no_curriculum_seed456",
      "arm": "max24_no_curriculum",
      "seed": 456,
      "stage": "stage1_max24",
      "stage_idx": 1,
      "local_step": 450,
      "global_step": 450,
      "compiler_max_steps": 24,
      "elapsed_sec": 1538.12,
      "expansion_event": "init",
      "init_loss": 0.5641363859176636,
      "op_loss": 1.3849110473529436e-05,
      "arg_loss": 0.010739540681242943,
      "state_loss": 0.9233291745185852,
      "state_train_accuracy": 0.7983193397521973,
      "executor_loss": 0.6429696679115295,
      "loss": 3.865816354751587
    },
    {
      "run": "main_max24_no_curriculum_seed456",
      "arm": "max24_no_curriculum",
      "seed": 456,
      "stage": "stage1_max24",
      "stage_idx": 1,
      "local_step": 500,
      "global_step": 500,
      "compiler_max_steps": 24,
      "elapsed_sec": 1709.949,
      "expansion_event": "init",
      "init_loss": 0.005333948880434036,
      "op_loss": 9.07163048395887e-05,
      "arg_loss": 0.00044877981417812407,
      "state_loss": 0.00998283363878727,
      "state_train_accuracy": 1.0,
      "executor_loss": 0.013425881043076515,
      "loss": 0.04663034528493881
    },
    {
      "run": "main_max24_no_curriculum_seed456",
      "arm": "max24_no_curriculum",
      "seed": 456,
      "stage": "stage1_max24",
      "stage_idx": 1,
      "local_step": 550,
      "global_step": 550,
      "compiler_max_steps": 24,
      "elapsed_sec": 1881.708,
      "expansion_event": "init",
      "init_loss": 0.0017424402758479118,
      "op_loss": 0.04822991415858269,
      "arg_loss": 0.035915497690439224,
      "state_loss": 0.34479963779449463,
      "state_train_accuracy": 0.876288652420044,
      "executor_loss": 1.0171538591384888,
      "loss": 1.5608152151107788
    },
    {
      "run": "main_max24_no_curriculum_seed456",
      "arm": "max24_no_curriculum",
      "seed": 456,
      "stage": "stage1_max24",
      "stage_idx": 1,
      "local_step": 600,
      "global_step": 600,
      "compiler_max_steps": 24,
      "elapsed_sec": 2052.984,
      "expansion_event": "init",
      "init_loss": 0.0008090937044471502,
      "op_loss": 1.257083840755513e-05,
      "arg_loss": 0.00011136293323943391,
      "state_loss": 0.001861574244685471,
      "state_train_accuracy": 0.9999999403953552,
      "executor_loss": 0.0026835990138351917,
      "loss": 0.008239571005105972
    },
    {
      "run": "main_max24_no_curriculum_seed456",
      "arm": "max24_no_curriculum",
      "seed": 456,
      "stage": "stage1_max24",
      "stage_idx": 1,
      "local_step": 650,
      "global_step": 650,
      "compiler_max_steps": 24,
      "elapsed_sec": 2224.578,
      "expansion_event": "init",
      "init_loss": 0.0006119837053120136,
      "op_loss": 3.61768843504251e-06,
      "arg_loss": 0.00013163179391995072,
      "state_loss": 0.0014759640907868743,
      "state_train_accuracy": 1.0,
      "executor_loss": 0.0026061313692480326,
      "loss": 0.00706017529591918
    },
    {
      "run": "main_max24_no_curriculum_seed456",
      "arm": "max24_no_curriculum",
      "seed": 456,
      "stage": "stage1_max24",
      "stage_idx": 1,
      "local_step": 700,
      "global_step": 700,
      "compiler_max_steps": 24,
      "elapsed_sec": 2394.501,
      "expansion_event": "init",
      "init_loss": 0.00036694412119686604,
      "op_loss": 0.11490454524755478,
      "arg_loss": 0.09451239556074142,
      "state_loss": 0.1643122434616089,
      "state_train_accuracy": 0.9767441749572754,
      "executor_loss": 1.24561607837677,
      "loss": 1.9043501615524292
    },
    {
      "run": "main_max24_no_curriculum_seed456",
      "arm": "max24_no_curriculum",
      "seed": 456,
      "stage": "stage1_max24",
      "stage_idx": 1,
      "local_step": 750,
      "global_step": 750,
      "compiler_max_steps": 24,
      "elapsed_sec": 2565.579,
      "expansion_event": "init",
      "init_loss": 0.00035451678559184074,
      "op_loss": 3.1450872484128922e-06,
      "arg_loss": 2.9508655643439852e-05,
      "state_loss": 0.0007130831945687532,
      "state_train_accuracy": 0.9999999403953552,
      "executor_loss": 0.00073830975452438,
      "loss": 0.0029906397685408592
    }
  ],
  "metrics": [
    {
      "stage": "stage1_max24",
      "split": "standard_L8",
      "n": 64,
      "max_steps": 24,
      "executor_accuracy": 1.0,
      "executor_target_mass": 0.999413512647152,
      "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_seed456",
      "arm": "max24_no_curriculum",
      "seed": 456,
      "global_step": 750,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "paraphrase_L8",
      "n": 64,
      "max_steps": 24,
      "executor_accuracy": 1.0,
      "executor_target_mass": 0.9993921220302582,
      "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_seed456",
      "arm": "max24_no_curriculum",
      "seed": 456,
      "global_step": 750,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "heldout_L8",
      "n": 64,
      "max_steps": 24,
      "executor_accuracy": 1.0,
      "executor_target_mass": 0.99793840944767,
      "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_seed456",
      "arm": "max24_no_curriculum",
      "seed": 456,
      "global_step": 750,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "paired_L8",
      "n": 64,
      "max_steps": 24,
      "executor_accuracy": 1.0,
      "executor_target_mass": 0.9993906766176224,
      "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_seed456",
      "arm": "max24_no_curriculum",
      "seed": 456,
      "global_step": 750,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "paired_heldout_L8",
      "n": 64,
      "max_steps": 24,
      "executor_accuracy": 1.0,
      "executor_target_mass": 0.9985420107841492,
      "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_seed456",
      "arm": "max24_no_curriculum",
      "seed": 456,
      "global_step": 750,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "standard_L16",
      "n": 64,
      "max_steps": 24,
      "executor_accuracy": 1.0,
      "executor_target_mass": 0.9991767331957817,
      "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": 16.0,
      "run": "main_max24_no_curriculum_seed456",
      "arm": "max24_no_curriculum",
      "seed": 456,
      "global_step": 750,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "paraphrase_L16",
      "n": 64,
      "max_steps": 24,
      "executor_accuracy": 0.9375,
      "executor_target_mass": 0.9366795271635056,
      "direct_accuracy": NaN,
      "init_accuracy": 1.0,
      "op_accuracy": 0.9990234375,
      "arg_accuracy": 0.9970703125,
      "program_exact": 0.9375,
      "state_accuracy": 0.99609375,
      "state_all_exact": 0.9375,
      "state_prefix_fraction": 0.99609375,
      "state_mean_correct_prefix": 15.9375,
      "run": "main_max24_no_curriculum_seed456",
      "arm": "max24_no_curriculum",
      "seed": 456,
      "global_step": 750,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "heldout_L16",
      "n": 64,
      "max_steps": 24,
      "executor_accuracy": 0.75,
      "executor_target_mass": 0.7290827818214893,
      "direct_accuracy": NaN,
      "init_accuracy": 1.0,
      "op_accuracy": 0.98828125,
      "arg_accuracy": 0.9794921875,
      "program_exact": 0.75,
      "state_accuracy": 0.9638671875,
      "state_all_exact": 0.75,
      "state_prefix_fraction": 0.9638671875,
      "state_mean_correct_prefix": 15.421875,
      "run": "main_max24_no_curriculum_seed456",
      "arm": "max24_no_curriculum",
      "seed": 456,
      "global_step": 750,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "paired_L16",
      "n": 64,
      "max_steps": 24,
      "executor_accuracy": 0.96875,
      "executor_target_mass": 0.9679021239280701,
      "direct_accuracy": NaN,
      "init_accuracy": 1.0,
      "op_accuracy": 0.998046875,
      "arg_accuracy": 0.998046875,
      "program_exact": 0.96875,
      "state_accuracy": 0.998046875,
      "state_all_exact": 0.96875,
      "state_prefix_fraction": 0.998046875,
      "state_mean_correct_prefix": 15.96875,
      "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_seed456",
      "arm": "max24_no_curriculum",
      "seed": 456,
      "global_step": 750,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "paired_heldout_L16",
      "n": 64,
      "max_steps": 24,
      "executor_accuracy": 0.828125,
      "executor_target_mass": 0.8238180801272392,
      "direct_accuracy": NaN,
      "init_accuracy": 1.0,
      "op_accuracy": 0.9912109375,
      "arg_accuracy": 0.9892578125,
      "program_exact": 0.828125,
      "state_accuracy": 0.98828125,
      "state_all_exact": 0.828125,
      "state_prefix_fraction": 0.98828125,
      "state_mean_correct_prefix": 15.8125,
      "executor_pair_answer_consistency": 0.65625,
      "executor_pair_both_correct": 0.65625,
      "compiler_pair_program_consistency": 0.65625,
      "compiler_pair_state_consistency": 0.65625,
      "run": "main_max24_no_curriculum_seed456",
      "arm": "max24_no_curriculum",
      "seed": 456,
      "global_step": 750,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "standard_L24",
      "n": 64,
      "max_steps": 24,
      "executor_accuracy": 0.0,
      "executor_target_mass": 5.007966626635607e-05,
      "direct_accuracy": NaN,
      "init_accuracy": 1.0,
      "op_accuracy": 0.884765625,
      "arg_accuracy": 0.8307291666666666,
      "program_exact": 0.0,
      "state_accuracy": 0.791015625,
      "state_all_exact": 0.0,
      "state_prefix_fraction": 0.7903645833333334,
      "state_mean_correct_prefix": 18.96875,
      "run": "main_max24_no_curriculum_seed456",
      "arm": "max24_no_curriculum",
      "seed": 456,
      "global_step": 750,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "paraphrase_L24",
      "n": 64,
      "max_steps": 24,
      "executor_accuracy": 1.0,
      "executor_target_mass": 0.996421605348587,
      "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": 24.0,
      "run": "main_max24_no_curriculum_seed456",
      "arm": "max24_no_curriculum",
      "seed": 456,
      "global_step": 750,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "heldout_L24",
      "n": 64,
      "max_steps": 24,
      "executor_accuracy": 0.375,
      "executor_target_mass": 0.37129905069468805,
      "direct_accuracy": NaN,
      "init_accuracy": 1.0,
      "op_accuracy": 0.919921875,
      "arg_accuracy": 0.8782552083333334,
      "program_exact": 0.375,
      "state_accuracy": 0.7916666666666666,
      "state_all_exact": 0.375,
      "state_prefix_fraction": 0.7916666666666666,
      "state_mean_correct_prefix": 19.0,
      "run": "main_max24_no_curriculum_seed456",
      "arm": "max24_no_curriculum",
      "seed": 456,
      "global_step": 750,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "paired_L24",
      "n": 64,
      "max_steps": 24,
      "executor_accuracy": 0.5,
      "executor_target_mass": 0.5034634135663509,
      "direct_accuracy": NaN,
      "init_accuracy": 1.0,
      "op_accuracy": 0.9401041666666666,
      "arg_accuracy": 0.916015625,
      "program_exact": 0.5,
      "state_accuracy": 0.892578125,
      "state_all_exact": 0.5,
      "state_prefix_fraction": 0.8919270833333334,
      "state_mean_correct_prefix": 21.40625,
      "executor_pair_answer_consistency": 0.03125,
      "executor_pair_both_correct": 0.03125,
      "compiler_pair_program_consistency": 0.03125,
      "compiler_pair_state_consistency": 0.03125,
      "run": "main_max24_no_curriculum_seed456",
      "arm": "max24_no_curriculum",
      "seed": 456,
      "global_step": 750,
      "stage_idx": 1
    },
    {
      "stage": "stage1_max24",
      "split": "paired_heldout_L24",
      "n": 64,
      "max_steps": 24,
      "executor_accuracy": 0.25,
      "executor_target_mass": 0.24758669227048813,
      "direct_accuracy": NaN,
      "init_accuracy": 1.0,
      "op_accuracy": 0.9088541666666666,
      "arg_accuracy": 0.8723958333333334,
      "program_exact": 0.25,
      "state_accuracy": 0.8333333333333334,
      "state_all_exact": 0.25,
      "state_prefix_fraction": 0.8326822916666666,
      "state_mean_correct_prefix": 19.984375,
      "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_seed456",
      "arm": "max24_no_curriculum",
      "seed": 456,
      "global_step": 750,
      "stage_idx": 1
    }
  ],
  "checkpoints": [
    {
      "run": "main_max24_no_curriculum_seed456",
      "seed": 456,
      "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_seed456/stage1_max24_step00750",
      "model_checkpoint": "/workspace/large_artifacts/qwen_compiler_multiseed_reattribution/checkpoints/main_max24_no_curriculum_seed456/stage1_max24_step00750/adapter",
      "heads_checkpoint": "/workspace/large_artifacts/qwen_compiler_multiseed_reattribution/checkpoints/main_max24_no_curriculum_seed456/stage1_max24_step00750/heads.pt",
      "selection_split": "paired_L24",
      "selection_metric": "executor_pair_both_correct",
      "selection_value": 0.03125,
      "executor_accuracy": 0.5,
      "program_exact": 0.5,
      "state_prefix_fraction": 0.8919270833333334
    }
  ],
  "elapsed_sec": 2692.679
}