{
  "args": {
    "model_id": "Qwen/Qwen3.5-4B",
    "text_model_fallback": "Qwen/Qwen3-4B",
    "model_class": "auto",
    "load_in_4bit": true,
    "torch_dtype": "bf16",
    "device_map": "auto",
    "modulus": 97,
    "max_steps": 24,
    "train_min_len": 1,
    "train_max_len": 4,
    "answer_train_min_len": 1,
    "answer_train_max_len": 8,
    "eval_lengths": "4,8,12,24",
    "train_size": 1024,
    "answer_train_size": 1024,
    "eval_size": 256,
    "feature_batch_size": 8,
    "train_batch_size": 128,
    "eval_batch_size": 256,
    "bootstrap_steps": 800,
    "answer_steps": 800,
    "eval_every": 400,
    "stage_eval_every": 400,
    "lr": 0.001,
    "answer_lr_scale": 0.1,
    "weight_decay": 0.01,
    "grad_clip": 1.0,
    "head_width": 512,
    "trace_loss_weight": 1.0,
    "executor_loss_weight": 1.0,
    "variants": "direct,compiler_trace,compiler_answer_only,compiler_trace_then_answer,compiler_trace_then_answer_low_lr",
    "max_length": 512,
    "seed": 7,
    "output_dir": "experiments/qwen_trace_bootstrap_retention/runs/main_qwen35_retention",
    "checkpoint_dir": "",
    "verbose": false
  },
  "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": "AutoModelForMultimodalLM",
    "peft_installed": true,
    "gpu_name": "NVIDIA RTX 6000 Ada Generation",
    "gpu_vram_gb": 47.383
  },
  "hidden_dim": 2560,
  "dataset": {
    "bootstrap_train_size": 1024,
    "answer_train_size": 1024,
    "eval_size": 256,
    "bootstrap_train_lengths": [
      1,
      4
    ],
    "answer_train_lengths": [
      1,
      8
    ],
    "eval_lengths": [
      4,
      8,
      12,
      24
    ]
  },
  "variants": {
    "direct": {
      "variant": "direct",
      "train_seconds": 69.45988416671753,
      "checkpoints": [
        "large_artifacts/qwen_trace_bootstrap_retention/checkpoints/main_qwen35_retention/direct/checkpoint_final.pt"
      ],
      "train_log": [
        {
          "stage": "answer_only",
          "local_step": 1,
          "step": 1,
          "loss": 4.581656455993652,
          "trace_loss_active": false,
          "lr_scale": 1.0,
          "direct_loss": 4.581656455993652,
          "len4_direct_accuracy": 0.0078125,
          "len4_direct_target_mass": 0.009281355887651443,
          "len4_executor_accuracy": NaN,
          "len4_executor_target_mass": NaN,
          "len4_init_accuracy": NaN,
          "len4_op_accuracy": NaN,
          "len4_arg_accuracy": NaN,
          "len4_program_exact": NaN,
          "len8_direct_accuracy": 0.0078125,
          "len8_direct_target_mass": 0.009220581501722336,
          "len8_executor_accuracy": NaN,
          "len8_executor_target_mass": NaN,
          "len8_init_accuracy": NaN,
          "len8_op_accuracy": NaN,
          "len8_arg_accuracy": NaN,
          "len8_program_exact": NaN,
          "len12_direct_accuracy": 0.015625,
          "len12_direct_target_mass": 0.01212192140519619,
          "len12_executor_accuracy": NaN,
          "len12_executor_target_mass": NaN,
          "len12_init_accuracy": NaN,
          "len12_op_accuracy": NaN,
          "len12_arg_accuracy": NaN,
          "len12_program_exact": NaN,
          "len24_direct_accuracy": 0.00390625,
          "len24_direct_target_mass": 0.009910820052027702,
          "len24_executor_accuracy": NaN,
          "len24_executor_target_mass": NaN,
          "len24_init_accuracy": NaN,
          "len24_op_accuracy": NaN,
          "len24_arg_accuracy": NaN,
          "len24_program_exact": NaN
        },
        {
          "stage": "answer_only",
          "local_step": 400,
          "step": 400,
          "loss": 4.323907852172852,
          "trace_loss_active": false,
          "lr_scale": 1.0,
          "direct_loss": 4.323907852172852,
          "len4_direct_accuracy": 0.015625,
          "len4_direct_target_mass": 0.011056570336222649,
          "len4_executor_accuracy": NaN,
          "len4_executor_target_mass": NaN,
          "len4_init_accuracy": NaN,
          "len4_op_accuracy": NaN,
          "len4_arg_accuracy": NaN,
          "len4_program_exact": NaN,
          "len8_direct_accuracy": 0.00390625,
          "len8_direct_target_mass": 0.009859984740614891,
          "len8_executor_accuracy": NaN,
          "len8_executor_target_mass": NaN,
          "len8_init_accuracy": NaN,
          "len8_op_accuracy": NaN,
          "len8_arg_accuracy": NaN,
          "len8_program_exact": NaN,
          "len12_direct_accuracy": 0.0,
          "len12_direct_target_mass": 0.009882545098662376,
          "len12_executor_accuracy": NaN,
          "len12_executor_target_mass": NaN,
          "len12_init_accuracy": NaN,
          "len12_op_accuracy": NaN,
          "len12_arg_accuracy": NaN,
          "len12_program_exact": NaN,
          "len24_direct_accuracy": 0.00390625,
          "len24_direct_target_mass": 0.010453540831804276,
          "len24_executor_accuracy": NaN,
          "len24_executor_target_mass": NaN,
          "len24_init_accuracy": NaN,
          "len24_op_accuracy": NaN,
          "len24_arg_accuracy": NaN,
          "len24_program_exact": NaN
        },
        {
          "stage": "answer_only",
          "local_step": 800,
          "step": 800,
          "loss": 4.086009979248047,
          "trace_loss_active": false,
          "lr_scale": 1.0,
          "direct_loss": 4.086009979248047,
          "len4_direct_accuracy": 0.0078125,
          "len4_direct_target_mass": 0.010806550271809101,
          "len4_executor_accuracy": NaN,
          "len4_executor_target_mass": NaN,
          "len4_init_accuracy": NaN,
          "len4_op_accuracy": NaN,
          "len4_arg_accuracy": NaN,
          "len4_program_exact": NaN,
          "len8_direct_accuracy": 0.015625,
          "len8_direct_target_mass": 0.009627658873796463,
          "len8_executor_accuracy": NaN,
          "len8_executor_target_mass": NaN,
          "len8_init_accuracy": NaN,
          "len8_op_accuracy": NaN,
          "len8_arg_accuracy": NaN,
          "len8_program_exact": NaN,
          "len12_direct_accuracy": 0.00390625,
          "len12_direct_target_mass": 0.010019897483289242,
          "len12_executor_accuracy": NaN,
          "len12_executor_target_mass": NaN,
          "len12_init_accuracy": NaN,
          "len12_op_accuracy": NaN,
          "len12_arg_accuracy": NaN,
          "len12_program_exact": NaN,
          "len24_direct_accuracy": 0.0078125,
          "len24_direct_target_mass": 0.01089046336710453,
          "len24_executor_accuracy": NaN,
          "len24_executor_target_mass": NaN,
          "len24_init_accuracy": NaN,
          "len24_op_accuracy": NaN,
          "len24_arg_accuracy": NaN,
          "len24_program_exact": NaN
        },
        {
          "stage": "answer_only",
          "local_step": 1200,
          "step": 1200,
          "loss": 3.8974480628967285,
          "trace_loss_active": false,
          "lr_scale": 1.0,
          "direct_loss": 3.8974480628967285,
          "len4_direct_accuracy": 0.015625,
          "len4_direct_target_mass": 0.010973562486469746,
          "len4_executor_accuracy": NaN,
          "len4_executor_target_mass": NaN,
          "len4_init_accuracy": NaN,
          "len4_op_accuracy": NaN,
          "len4_arg_accuracy": NaN,
          "len4_program_exact": NaN,
          "len8_direct_accuracy": 0.0078125,
          "len8_direct_target_mass": 0.009822966530919075,
          "len8_executor_accuracy": NaN,
          "len8_executor_target_mass": NaN,
          "len8_init_accuracy": NaN,
          "len8_op_accuracy": NaN,
          "len8_arg_accuracy": NaN,
          "len8_program_exact": NaN,
          "len12_direct_accuracy": 0.0,
          "len12_direct_target_mass": 0.010114362463355064,
          "len12_executor_accuracy": NaN,
          "len12_executor_target_mass": NaN,
          "len12_init_accuracy": NaN,
          "len12_op_accuracy": NaN,
          "len12_arg_accuracy": NaN,
          "len12_program_exact": NaN,
          "len24_direct_accuracy": 0.015625,
          "len24_direct_target_mass": 0.010277463123202324,
          "len24_executor_accuracy": NaN,
          "len24_executor_target_mass": NaN,
          "len24_init_accuracy": NaN,
          "len24_op_accuracy": NaN,
          "len24_arg_accuracy": NaN,
          "len24_program_exact": NaN
        },
        {
          "stage": "answer_only",
          "local_step": 1600,
          "step": 1600,
          "loss": 3.601148843765259,
          "trace_loss_active": false,
          "lr_scale": 1.0,
          "direct_loss": 3.601148843765259,
          "len4_direct_accuracy": 0.0078125,
          "len4_direct_target_mass": 0.011590146459639072,
          "len4_executor_accuracy": NaN,
          "len4_executor_target_mass": NaN,
          "len4_init_accuracy": NaN,
          "len4_op_accuracy": NaN,
          "len4_arg_accuracy": NaN,
          "len4_program_exact": NaN,
          "len8_direct_accuracy": 0.0078125,
          "len8_direct_target_mass": 0.01051628403365612,
          "len8_executor_accuracy": NaN,
          "len8_executor_target_mass": NaN,
          "len8_init_accuracy": NaN,
          "len8_op_accuracy": NaN,
          "len8_arg_accuracy": NaN,
          "len8_program_exact": NaN,
          "len12_direct_accuracy": 0.01171875,
          "len12_direct_target_mass": 0.01033981516957283,
          "len12_executor_accuracy": NaN,
          "len12_executor_target_mass": NaN,
          "len12_init_accuracy": NaN,
          "len12_op_accuracy": NaN,
          "len12_arg_accuracy": NaN,
          "len12_program_exact": NaN,
          "len24_direct_accuracy": 0.00390625,
          "len24_direct_target_mass": 0.009672501124441624,
          "len24_executor_accuracy": NaN,
          "len24_executor_target_mass": NaN,
          "len24_init_accuracy": NaN,
          "len24_op_accuracy": NaN,
          "len24_arg_accuracy": NaN,
          "len24_program_exact": NaN
        }
      ],
      "final_metrics": {
        "len4": {
          "variant": "direct",
          "n": 256.0,
          "direct_accuracy": 0.0078125,
          "direct_target_mass": 0.011590146459639072,
          "executor_accuracy": NaN,
          "executor_target_mass": NaN,
          "init_accuracy": NaN,
          "op_accuracy": NaN,
          "arg_accuracy": NaN,
          "program_exact": NaN
        },
        "len8": {
          "variant": "direct",
          "n": 256.0,
          "direct_accuracy": 0.0078125,
          "direct_target_mass": 0.01051628403365612,
          "executor_accuracy": NaN,
          "executor_target_mass": NaN,
          "init_accuracy": NaN,
          "op_accuracy": NaN,
          "arg_accuracy": NaN,
          "program_exact": NaN
        },
        "len12": {
          "variant": "direct",
          "n": 256.0,
          "direct_accuracy": 0.01171875,
          "direct_target_mass": 0.01033981516957283,
          "executor_accuracy": NaN,
          "executor_target_mass": NaN,
          "init_accuracy": NaN,
          "op_accuracy": NaN,
          "arg_accuracy": NaN,
          "program_exact": NaN
        },
        "len24": {
          "variant": "direct",
          "n": 256.0,
          "direct_accuracy": 0.00390625,
          "direct_target_mass": 0.009672501124441624,
          "executor_accuracy": NaN,
          "executor_target_mass": NaN,
          "init_accuracy": NaN,
          "op_accuracy": NaN,
          "arg_accuracy": NaN,
          "program_exact": NaN
        }
      }
    },
    "compiler_trace": {
      "variant": "compiler_trace",
      "train_seconds": 104.02752733230591,
      "checkpoints": [
        "large_artifacts/qwen_trace_bootstrap_retention/checkpoints/main_qwen35_retention/compiler_trace/checkpoint_final.pt"
      ],
      "train_log": [
        {
          "stage": "trace_bootstrap",
          "local_step": 1,
          "step": 1,
          "loss": 14.88444995880127,
          "trace_loss_active": true,
          "lr_scale": 1.0,
          "executor_loss": 4.577169418334961,
          "init_loss": 4.602089881896973,
          "op_loss": 1.1234331130981445,
          "arg_loss": 4.581757545471191,
          "len4_direct_accuracy": NaN,
          "len4_direct_target_mass": NaN,
          "len4_executor_accuracy": 0.0078125,
          "len4_executor_target_mass": 0.010296441614627838,
          "len4_init_accuracy": 0.0234375,
          "len4_op_accuracy": 0.4111328125,
          "len4_arg_accuracy": 0.0888671875,
          "len4_program_exact": 0.0,
          "len8_direct_accuracy": NaN,
          "len8_direct_target_mass": NaN,
          "len8_executor_accuracy": 0.0078125,
          "len8_executor_target_mass": 0.010308479890227318,
          "len8_init_accuracy": 0.0390625,
          "len8_op_accuracy": 0.38427734375,
          "len8_arg_accuracy": 0.07421875,
          "len8_program_exact": 0.0,
          "len12_direct_accuracy": NaN,
          "len12_direct_target_mass": NaN,
          "len12_executor_accuracy": 0.01953125,
          "len12_executor_target_mass": 0.010306254029273987,
          "len12_init_accuracy": 0.03515625,
          "len12_op_accuracy": 0.3704427083333333,
          "len12_arg_accuracy": 0.08658854166666667,
          "len12_program_exact": 0.0,
          "len24_direct_accuracy": NaN,
          "len24_direct_target_mass": NaN,
          "len24_executor_accuracy": 0.01171875,
          "len24_executor_target_mass": 0.010304299183189869,
          "len24_init_accuracy": 0.04296875,
          "len24_op_accuracy": 0.35498046875,
          "len24_arg_accuracy": 0.08170572916666667,
          "len24_program_exact": 0.0
        },
        {
          "stage": "trace_bootstrap",
          "local_step": 400,
          "step": 400,
          "loss": 0.0008173511596396565,
          "trace_loss_active": true,
          "lr_scale": 1.0,
          "executor_loss": 0.00046962127089500427,
          "init_loss": 0.0002744930097833276,
          "op_loss": 1.786903158063069e-05,
          "arg_loss": 5.536782555282116e-05,
          "len4_direct_accuracy": NaN,
          "len4_direct_target_mass": NaN,
          "len4_executor_accuracy": 1.0,
          "len4_executor_target_mass": 0.9974122643470764,
          "len4_init_accuracy": 1.0,
          "len4_op_accuracy": 1.0,
          "len4_arg_accuracy": 1.0,
          "len4_program_exact": 1.0,
          "len8_direct_accuracy": NaN,
          "len8_direct_target_mass": NaN,
          "len8_executor_accuracy": 0.984375,
          "len8_executor_target_mass": 0.9821889400482178,
          "len8_init_accuracy": 1.0,
          "len8_op_accuracy": 0.998046875,
          "len8_arg_accuracy": 1.0,
          "len8_program_exact": 0.984375,
          "len12_direct_accuracy": NaN,
          "len12_direct_target_mass": NaN,
          "len12_executor_accuracy": 0.93359375,
          "len12_executor_target_mass": 0.9300813674926758,
          "len12_init_accuracy": 1.0,
          "len12_op_accuracy": 0.994140625,
          "len12_arg_accuracy": 1.0,
          "len12_program_exact": 0.93359375,
          "len24_direct_accuracy": NaN,
          "len24_direct_target_mass": NaN,
          "len24_executor_accuracy": 0.8515625,
          "len24_executor_target_mass": 0.815548837184906,
          "len24_init_accuracy": 1.0,
          "len24_op_accuracy": 0.9931640625,
          "len24_arg_accuracy": 1.0,
          "len24_program_exact": 0.8515625
        },
        {
          "stage": "trace_bootstrap",
          "local_step": 800,
          "step": 800,
          "loss": 0.0002411853929515928,
          "trace_loss_active": true,
          "lr_scale": 1.0,
          "executor_loss": 0.00013204297283664346,
          "init_loss": 9.276399214286357e-05,
          "op_loss": 3.926977569790324e-06,
          "arg_loss": 1.2451453585526906e-05,
          "len4_direct_accuracy": NaN,
          "len4_direct_target_mass": NaN,
          "len4_executor_accuracy": 0.99609375,
          "len4_executor_target_mass": 0.9975165724754333,
          "len4_init_accuracy": 1.0,
          "len4_op_accuracy": 0.9990234375,
          "len4_arg_accuracy": 1.0,
          "len4_program_exact": 0.99609375,
          "len8_direct_accuracy": NaN,
          "len8_direct_target_mass": NaN,
          "len8_executor_accuracy": 0.984375,
          "len8_executor_target_mass": 0.9840989708900452,
          "len8_init_accuracy": 1.0,
          "len8_op_accuracy": 0.998046875,
          "len8_arg_accuracy": 1.0,
          "len8_program_exact": 0.984375,
          "len12_direct_accuracy": NaN,
          "len12_direct_target_mass": NaN,
          "len12_executor_accuracy": 0.9375,
          "len12_executor_target_mass": 0.9322890043258667,
          "len12_init_accuracy": 1.0,
          "len12_op_accuracy": 0.9944661458333334,
          "len12_arg_accuracy": 1.0,
          "len12_program_exact": 0.9375,
          "len24_direct_accuracy": NaN,
          "len24_direct_target_mass": NaN,
          "len24_executor_accuracy": 0.8515625,
          "len24_executor_target_mass": 0.8241989612579346,
          "len24_init_accuracy": 1.0,
          "len24_op_accuracy": 0.9930013020833334,
          "len24_arg_accuracy": 1.0,
          "len24_program_exact": 0.8515625
        },
        {
          "stage": "trace_continuation",
          "local_step": 1,
          "step": 801,
          "loss": 0.010952366515994072,
          "trace_loss_active": true,
          "lr_scale": 1.0,
          "executor_loss": 0.008861883543431759,
          "init_loss": 0.00011607771011767909,
          "op_loss": 0.001953862374648452,
          "arg_loss": 2.0542425772873685e-05,
          "len4_direct_accuracy": NaN,
          "len4_direct_target_mass": NaN,
          "len4_executor_accuracy": 0.99609375,
          "len4_executor_target_mass": 0.9932102560997009,
          "len4_init_accuracy": 1.0,
          "len4_op_accuracy": 0.9990234375,
          "len4_arg_accuracy": 1.0,
          "len4_program_exact": 0.99609375,
          "len8_direct_accuracy": NaN,
          "len8_direct_target_mass": NaN,
          "len8_executor_accuracy": 0.99609375,
          "len8_executor_target_mass": 0.9867865443229675,
          "len8_init_accuracy": 1.0,
          "len8_op_accuracy": 0.99951171875,
          "len8_arg_accuracy": 1.0,
          "len8_program_exact": 0.99609375,
          "len12_direct_accuracy": NaN,
          "len12_direct_target_mass": NaN,
          "len12_executor_accuracy": 0.94921875,
          "len12_executor_target_mass": 0.935314953327179,
          "len12_init_accuracy": 1.0,
          "len12_op_accuracy": 0.9957682291666666,
          "len12_arg_accuracy": 1.0,
          "len12_program_exact": 0.94921875,
          "len24_direct_accuracy": NaN,
          "len24_direct_target_mass": NaN,
          "len24_executor_accuracy": 0.8984375,
          "len24_executor_target_mass": 0.8544360399246216,
          "len24_init_accuracy": 1.0,
          "len24_op_accuracy": 0.9954427083333334,
          "len24_arg_accuracy": 1.0,
          "len24_program_exact": 0.8984375
        },
        {
          "stage": "trace_continuation",
          "local_step": 400,
          "step": 1200,
          "loss": 0.00012387585593387485,
          "trace_loss_active": true,
          "lr_scale": 1.0,
          "executor_loss": 7.183942943811417e-05,
          "init_loss": 4.632883792510256e-05,
          "op_loss": 4.5224453515402274e-08,
          "arg_loss": 5.662363037117757e-06,
          "len4_direct_accuracy": NaN,
          "len4_direct_target_mass": NaN,
          "len4_executor_accuracy": 1.0,
          "len4_executor_target_mass": 0.9997814297676086,
          "len4_init_accuracy": 1.0,
          "len4_op_accuracy": 1.0,
          "len4_arg_accuracy": 1.0,
          "len4_program_exact": 1.0,
          "len8_direct_accuracy": NaN,
          "len8_direct_target_mass": NaN,
          "len8_executor_accuracy": 1.0,
          "len8_executor_target_mass": 0.999042809009552,
          "len8_init_accuracy": 1.0,
          "len8_op_accuracy": 1.0,
          "len8_arg_accuracy": 1.0,
          "len8_program_exact": 1.0,
          "len12_direct_accuracy": NaN,
          "len12_direct_target_mass": NaN,
          "len12_executor_accuracy": 0.9921875,
          "len12_executor_target_mass": 0.9889588356018066,
          "len12_init_accuracy": 1.0,
          "len12_op_accuracy": 0.9993489583333334,
          "len12_arg_accuracy": 1.0,
          "len12_program_exact": 0.9921875,
          "len24_direct_accuracy": NaN,
          "len24_direct_target_mass": NaN,
          "len24_executor_accuracy": 0.94921875,
          "len24_executor_target_mass": 0.9366211295127869,
          "len24_init_accuracy": 1.0,
          "len24_op_accuracy": 0.9977213541666666,
          "len24_arg_accuracy": 1.0,
          "len24_program_exact": 0.94921875
        },
        {
          "stage": "trace_continuation",
          "local_step": 800,
          "step": 1600,
          "loss": 8.193748362828046e-05,
          "trace_loss_active": true,
          "lr_scale": 1.0,
          "executor_loss": 4.790968159795739e-05,
          "init_loss": 3.0336253985296935e-05,
          "op_loss": 2.8238449090167705e-07,
          "arg_loss": 3.4091597171936883e-06,
          "len4_direct_accuracy": NaN,
          "len4_direct_target_mass": NaN,
          "len4_executor_accuracy": 1.0,
          "len4_executor_target_mass": 0.999833881855011,
          "len4_init_accuracy": 1.0,
          "len4_op_accuracy": 1.0,
          "len4_arg_accuracy": 1.0,
          "len4_program_exact": 1.0,
          "len8_direct_accuracy": NaN,
          "len8_direct_target_mass": NaN,
          "len8_executor_accuracy": 1.0,
          "len8_executor_target_mass": 0.9988480806350708,
          "len8_init_accuracy": 1.0,
          "len8_op_accuracy": 1.0,
          "len8_arg_accuracy": 1.0,
          "len8_program_exact": 1.0,
          "len12_direct_accuracy": NaN,
          "len12_direct_target_mass": NaN,
          "len12_executor_accuracy": 0.9921875,
          "len12_executor_target_mass": 0.9894802570343018,
          "len12_init_accuracy": 1.0,
          "len12_op_accuracy": 0.9993489583333334,
          "len12_arg_accuracy": 1.0,
          "len12_program_exact": 0.9921875,
          "len24_direct_accuracy": NaN,
          "len24_direct_target_mass": NaN,
          "len24_executor_accuracy": 0.9609375,
          "len24_executor_target_mass": 0.9415032863616943,
          "len24_init_accuracy": 1.0,
          "len24_op_accuracy": 0.9982096354166666,
          "len24_arg_accuracy": 1.0,
          "len24_program_exact": 0.9609375
        }
      ],
      "final_metrics": {
        "len4": {
          "variant": "compiler_trace",
          "n": 256.0,
          "direct_accuracy": NaN,
          "direct_target_mass": NaN,
          "executor_accuracy": 1.0,
          "executor_target_mass": 0.999833881855011,
          "init_accuracy": 1.0,
          "op_accuracy": 1.0,
          "arg_accuracy": 1.0,
          "program_exact": 1.0
        },
        "len8": {
          "variant": "compiler_trace",
          "n": 256.0,
          "direct_accuracy": NaN,
          "direct_target_mass": NaN,
          "executor_accuracy": 1.0,
          "executor_target_mass": 0.9988480806350708,
          "init_accuracy": 1.0,
          "op_accuracy": 1.0,
          "arg_accuracy": 1.0,
          "program_exact": 1.0
        },
        "len12": {
          "variant": "compiler_trace",
          "n": 256.0,
          "direct_accuracy": NaN,
          "direct_target_mass": NaN,
          "executor_accuracy": 0.9921875,
          "executor_target_mass": 0.9894802570343018,
          "init_accuracy": 1.0,
          "op_accuracy": 0.9993489583333334,
          "arg_accuracy": 1.0,
          "program_exact": 0.9921875
        },
        "len24": {
          "variant": "compiler_trace",
          "n": 256.0,
          "direct_accuracy": NaN,
          "direct_target_mass": NaN,
          "executor_accuracy": 0.9609375,
          "executor_target_mass": 0.9415032863616943,
          "init_accuracy": 1.0,
          "op_accuracy": 0.9982096354166666,
          "arg_accuracy": 1.0,
          "program_exact": 0.9609375
        }
      }
    },
    "compiler_answer_only": {
      "variant": "compiler_answer_only",
      "train_seconds": 101.65164709091187,
      "checkpoints": [
        "large_artifacts/qwen_trace_bootstrap_retention/checkpoints/main_qwen35_retention/compiler_answer_only/checkpoint_final.pt"
      ],
      "train_log": [
        {
          "stage": "answer_only",
          "local_step": 1,
          "step": 1,
          "loss": 4.5781354904174805,
          "trace_loss_active": false,
          "lr_scale": 1.0,
          "executor_loss": 4.5781354904174805,
          "len4_direct_accuracy": NaN,
          "len4_direct_target_mass": NaN,
          "len4_executor_accuracy": 0.0078125,
          "len4_executor_target_mass": 0.010309262201189995,
          "len4_init_accuracy": 0.015625,
          "len4_op_accuracy": 0.333984375,
          "len4_arg_accuracy": 0.0166015625,
          "len4_program_exact": 0.0,
          "len8_direct_accuracy": NaN,
          "len8_direct_target_mass": NaN,
          "len8_executor_accuracy": 0.00390625,
          "len8_executor_target_mass": 0.010309277102351189,
          "len8_init_accuracy": 0.0078125,
          "len8_op_accuracy": 0.33251953125,
          "len8_arg_accuracy": 0.015625,
          "len8_program_exact": 0.0,
          "len12_direct_accuracy": NaN,
          "len12_direct_target_mass": NaN,
          "len12_executor_accuracy": 0.01171875,
          "len12_executor_target_mass": 0.010309278964996338,
          "len12_init_accuracy": 0.0078125,
          "len12_op_accuracy": 0.3310546875,
          "len12_arg_accuracy": 0.015625,
          "len12_program_exact": 0.0,
          "len24_direct_accuracy": NaN,
          "len24_direct_target_mass": NaN,
          "len24_executor_accuracy": 0.01171875,
          "len24_executor_target_mass": 0.010309278964996338,
          "len24_init_accuracy": 0.00390625,
          "len24_op_accuracy": 0.3367513020833333,
          "len24_arg_accuracy": 0.017903645833333332,
          "len24_program_exact": 0.0
        },
        {
          "stage": "answer_only",
          "local_step": 400,
          "step": 400,
          "loss": 3.6662511825561523,
          "trace_loss_active": false,
          "lr_scale": 1.0,
          "executor_loss": 3.6662511825561523,
          "len4_direct_accuracy": NaN,
          "len4_direct_target_mass": NaN,
          "len4_executor_accuracy": 0.01171875,
          "len4_executor_target_mass": 0.009829241782426834,
          "len4_init_accuracy": 0.0078125,
          "len4_op_accuracy": 0.333984375,
          "len4_arg_accuracy": 0.0009765625,
          "len4_program_exact": 0.0,
          "len8_direct_accuracy": NaN,
          "len8_direct_target_mass": NaN,
          "len8_executor_accuracy": 0.0078125,
          "len8_executor_target_mass": 0.010025721043348312,
          "len8_init_accuracy": 0.01171875,
          "len8_op_accuracy": 0.33251953125,
          "len8_arg_accuracy": 0.0009765625,
          "len8_program_exact": 0.0,
          "len12_direct_accuracy": NaN,
          "len12_direct_target_mass": NaN,
          "len12_executor_accuracy": 0.03125,
          "len12_executor_target_mass": 0.010366233065724373,
          "len12_init_accuracy": 0.0078125,
          "len12_op_accuracy": 0.3310546875,
          "len12_arg_accuracy": 0.0009765625,
          "len12_program_exact": 0.0,
          "len24_direct_accuracy": NaN,
          "len24_direct_target_mass": NaN,
          "len24_executor_accuracy": 0.01953125,
          "len24_executor_target_mass": 0.010304777882993221,
          "len24_init_accuracy": 0.0078125,
          "len24_op_accuracy": 0.3367513020833333,
          "len24_arg_accuracy": 0.00048828125,
          "len24_program_exact": 0.0
        },
        {
          "stage": "answer_only",
          "local_step": 800,
          "step": 800,
          "loss": 2.595015525817871,
          "trace_loss_active": false,
          "lr_scale": 1.0,
          "executor_loss": 2.595015525817871,
          "len4_direct_accuracy": NaN,
          "len4_direct_target_mass": NaN,
          "len4_executor_accuracy": 0.015625,
          "len4_executor_target_mass": 0.012093560770154,
          "len4_init_accuracy": 0.0078125,
          "len4_op_accuracy": 0.333984375,
          "len4_arg_accuracy": 0.0,
          "len4_program_exact": 0.0,
          "len8_direct_accuracy": NaN,
          "len8_direct_target_mass": NaN,
          "len8_executor_accuracy": 0.0078125,
          "len8_executor_target_mass": 0.010259105823934078,
          "len8_init_accuracy": 0.01171875,
          "len8_op_accuracy": 0.33251953125,
          "len8_arg_accuracy": 0.0,
          "len8_program_exact": 0.0,
          "len12_direct_accuracy": NaN,
          "len12_direct_target_mass": NaN,
          "len12_executor_accuracy": 0.0078125,
          "len12_executor_target_mass": 0.010218373499810696,
          "len12_init_accuracy": 0.0078125,
          "len12_op_accuracy": 0.3310546875,
          "len12_arg_accuracy": 0.0003255208333333333,
          "len12_program_exact": 0.0,
          "len24_direct_accuracy": NaN,
          "len24_direct_target_mass": NaN,
          "len24_executor_accuracy": 0.015625,
          "len24_executor_target_mass": 0.010185414925217628,
          "len24_init_accuracy": 0.0078125,
          "len24_op_accuracy": 0.3367513020833333,
          "len24_arg_accuracy": 0.0,
          "len24_program_exact": 0.0
        },
        {
          "stage": "answer_only",
          "local_step": 1200,
          "step": 1200,
          "loss": 1.730298399925232,
          "trace_loss_active": false,
          "lr_scale": 1.0,
          "executor_loss": 1.730298399925232,
          "len4_direct_accuracy": NaN,
          "len4_direct_target_mass": NaN,
          "len4_executor_accuracy": 0.01953125,
          "len4_executor_target_mass": 0.011603450402617455,
          "len4_init_accuracy": 0.0078125,
          "len4_op_accuracy": 0.333984375,
          "len4_arg_accuracy": 0.0,
          "len4_program_exact": 0.0,
          "len8_direct_accuracy": NaN,
          "len8_direct_target_mass": NaN,
          "len8_executor_accuracy": 0.01171875,
          "len8_executor_target_mass": 0.010543545708060265,
          "len8_init_accuracy": 0.01171875,
          "len8_op_accuracy": 0.33251953125,
          "len8_arg_accuracy": 0.0,
          "len8_program_exact": 0.0,
          "len12_direct_accuracy": NaN,
          "len12_direct_target_mass": NaN,
          "len12_executor_accuracy": 0.0078125,
          "len12_executor_target_mass": 0.009349539875984192,
          "len12_init_accuracy": 0.0078125,
          "len12_op_accuracy": 0.3310546875,
          "len12_arg_accuracy": 0.0,
          "len12_program_exact": 0.0,
          "len24_direct_accuracy": NaN,
          "len24_direct_target_mass": NaN,
          "len24_executor_accuracy": 0.015625,
          "len24_executor_target_mass": 0.010158355347812176,
          "len24_init_accuracy": 0.0078125,
          "len24_op_accuracy": 0.3367513020833333,
          "len24_arg_accuracy": 0.0,
          "len24_program_exact": 0.0
        },
        {
          "stage": "answer_only",
          "local_step": 1600,
          "step": 1600,
          "loss": 1.176544189453125,
          "trace_loss_active": false,
          "lr_scale": 1.0,
          "executor_loss": 1.176544189453125,
          "len4_direct_accuracy": NaN,
          "len4_direct_target_mass": NaN,
          "len4_executor_accuracy": 0.0078125,
          "len4_executor_target_mass": 0.010112432762980461,
          "len4_init_accuracy": 0.0078125,
          "len4_op_accuracy": 0.333984375,
          "len4_arg_accuracy": 0.0,
          "len4_program_exact": 0.0,
          "len8_direct_accuracy": NaN,
          "len8_direct_target_mass": NaN,
          "len8_executor_accuracy": 0.01171875,
          "len8_executor_target_mass": 0.008407964371144772,
          "len8_init_accuracy": 0.01171875,
          "len8_op_accuracy": 0.33251953125,
          "len8_arg_accuracy": 0.00048828125,
          "len8_program_exact": 0.0,
          "len12_direct_accuracy": NaN,
          "len12_direct_target_mass": NaN,
          "len12_executor_accuracy": 0.01171875,
          "len12_executor_target_mass": 0.009183306246995926,
          "len12_init_accuracy": 0.0078125,
          "len12_op_accuracy": 0.3310546875,
          "len12_arg_accuracy": 0.0,
          "len12_program_exact": 0.0,
          "len24_direct_accuracy": NaN,
          "len24_direct_target_mass": NaN,
          "len24_executor_accuracy": 0.0078125,
          "len24_executor_target_mass": 0.009735402651131153,
          "len24_init_accuracy": 0.0078125,
          "len24_op_accuracy": 0.3367513020833333,
          "len24_arg_accuracy": 0.0,
          "len24_program_exact": 0.0
        }
      ],
      "final_metrics": {
        "len4": {
          "variant": "compiler_answer_only",
          "n": 256.0,
          "direct_accuracy": NaN,
          "direct_target_mass": NaN,
          "executor_accuracy": 0.0078125,
          "executor_target_mass": 0.010112432762980461,
          "init_accuracy": 0.0078125,
          "op_accuracy": 0.333984375,
          "arg_accuracy": 0.0,
          "program_exact": 0.0
        },
        "len8": {
          "variant": "compiler_answer_only",
          "n": 256.0,
          "direct_accuracy": NaN,
          "direct_target_mass": NaN,
          "executor_accuracy": 0.01171875,
          "executor_target_mass": 0.008407964371144772,
          "init_accuracy": 0.01171875,
          "op_accuracy": 0.33251953125,
          "arg_accuracy": 0.00048828125,
          "program_exact": 0.0
        },
        "len12": {
          "variant": "compiler_answer_only",
          "n": 256.0,
          "direct_accuracy": NaN,
          "direct_target_mass": NaN,
          "executor_accuracy": 0.01171875,
          "executor_target_mass": 0.009183306246995926,
          "init_accuracy": 0.0078125,
          "op_accuracy": 0.3310546875,
          "arg_accuracy": 0.0,
          "program_exact": 0.0
        },
        "len24": {
          "variant": "compiler_answer_only",
          "n": 256.0,
          "direct_accuracy": NaN,
          "direct_target_mass": NaN,
          "executor_accuracy": 0.0078125,
          "executor_target_mass": 0.009735402651131153,
          "init_accuracy": 0.0078125,
          "op_accuracy": 0.3367513020833333,
          "arg_accuracy": 0.0,
          "program_exact": 0.0
        }
      }
    },
    "compiler_trace_then_answer": {
      "variant": "compiler_trace_then_answer",
      "train_seconds": 100.51543831825256,
      "checkpoints": [
        "large_artifacts/qwen_trace_bootstrap_retention/checkpoints/main_qwen35_retention/compiler_trace_then_answer/checkpoint_final.pt"
      ],
      "train_log": [
        {
          "stage": "trace_bootstrap",
          "local_step": 1,
          "step": 1,
          "loss": 14.824259757995605,
          "trace_loss_active": true,
          "lr_scale": 1.0,
          "executor_loss": 4.579751014709473,
          "init_loss": 4.582459449768066,
          "op_loss": 1.09193754196167,
          "arg_loss": 4.570111274719238,
          "len4_direct_accuracy": NaN,
          "len4_direct_target_mass": NaN,
          "len4_executor_accuracy": 0.015625,
          "len4_executor_target_mass": 0.010310400277376175,
          "len4_init_accuracy": 0.02734375,
          "len4_op_accuracy": 0.310546875,
          "len4_arg_accuracy": 0.185546875,
          "len4_program_exact": 0.0,
          "len8_direct_accuracy": NaN,
          "len8_direct_target_mass": NaN,
          "len8_executor_accuracy": 0.0078125,
          "len8_executor_target_mass": 0.01030952762812376,
          "len8_init_accuracy": 0.03125,
          "len8_op_accuracy": 0.32861328125,
          "len8_arg_accuracy": 0.162109375,
          "len8_program_exact": 0.0,
          "len12_direct_accuracy": NaN,
          "len12_direct_target_mass": NaN,
          "len12_executor_accuracy": 0.0078125,
          "len12_executor_target_mass": 0.010309104807674885,
          "len12_init_accuracy": 0.0546875,
          "len12_op_accuracy": 0.3349609375,
          "len12_arg_accuracy": 0.1806640625,
          "len12_program_exact": 0.0,
          "len24_direct_accuracy": NaN,
          "len24_direct_target_mass": NaN,
          "len24_executor_accuracy": 0.015625,
          "len24_executor_target_mass": 0.01030734647065401,
          "len24_init_accuracy": 0.03515625,
          "len24_op_accuracy": 0.3317057291666667,
          "len24_arg_accuracy": 0.16715494791666666,
          "len24_program_exact": 0.0
        },
        {
          "stage": "trace_bootstrap",
          "local_step": 400,
          "step": 400,
          "loss": 0.0007649306789971888,
          "trace_loss_active": true,
          "lr_scale": 1.0,
          "executor_loss": 0.00042066589230671525,
          "init_loss": 0.00028347905026748776,
          "op_loss": 4.429532509675482e-06,
          "arg_loss": 5.635617708321661e-05,
          "len4_direct_accuracy": NaN,
          "len4_direct_target_mass": NaN,
          "len4_executor_accuracy": 1.0,
          "len4_executor_target_mass": 0.9976519346237183,
          "len4_init_accuracy": 1.0,
          "len4_op_accuracy": 1.0,
          "len4_arg_accuracy": 1.0,
          "len4_program_exact": 1.0,
          "len8_direct_accuracy": NaN,
          "len8_direct_target_mass": NaN,
          "len8_executor_accuracy": 0.98828125,
          "len8_executor_target_mass": 0.9837839603424072,
          "len8_init_accuracy": 1.0,
          "len8_op_accuracy": 0.99853515625,
          "len8_arg_accuracy": 1.0,
          "len8_program_exact": 0.98828125,
          "len12_direct_accuracy": NaN,
          "len12_direct_target_mass": NaN,
          "len12_executor_accuracy": 0.9375,
          "len12_executor_target_mass": 0.9317259192466736,
          "len12_init_accuracy": 1.0,
          "len12_op_accuracy": 0.9944661458333334,
          "len12_arg_accuracy": 1.0,
          "len12_program_exact": 0.9375,
          "len24_direct_accuracy": NaN,
          "len24_direct_target_mass": NaN,
          "len24_executor_accuracy": 0.87109375,
          "len24_executor_target_mass": 0.8214964270591736,
          "len24_init_accuracy": 1.0,
          "len24_op_accuracy": 0.9939778645833334,
          "len24_arg_accuracy": 1.0,
          "len24_program_exact": 0.87109375
        },
        {
          "stage": "trace_bootstrap",
          "local_step": 800,
          "step": 800,
          "loss": 0.00023410184076055884,
          "trace_loss_active": true,
          "lr_scale": 1.0,
          "executor_loss": 0.00012916112609673291,
          "init_loss": 8.826862904243171e-05,
          "op_loss": 1.0299484074494103e-06,
          "arg_loss": 1.56421465362655e-05,
          "len4_direct_accuracy": NaN,
          "len4_direct_target_mass": NaN,
          "len4_executor_accuracy": 1.0,
          "len4_executor_target_mass": 0.9981541633605957,
          "len4_init_accuracy": 1.0,
          "len4_op_accuracy": 1.0,
          "len4_arg_accuracy": 1.0,
          "len4_program_exact": 1.0,
          "len8_direct_accuracy": NaN,
          "len8_direct_target_mass": NaN,
          "len8_executor_accuracy": 0.98828125,
          "len8_executor_target_mass": 0.9865819215774536,
          "len8_init_accuracy": 1.0,
          "len8_op_accuracy": 0.99853515625,
          "len8_arg_accuracy": 1.0,
          "len8_program_exact": 0.98828125,
          "len12_direct_accuracy": NaN,
          "len12_direct_target_mass": NaN,
          "len12_executor_accuracy": 0.9375,
          "len12_executor_target_mass": 0.9375402927398682,
          "len12_init_accuracy": 1.0,
          "len12_op_accuracy": 0.9947916666666666,
          "len12_arg_accuracy": 1.0,
          "len12_program_exact": 0.9375,
          "len24_direct_accuracy": NaN,
          "len24_direct_target_mass": NaN,
          "len24_executor_accuracy": 0.87109375,
          "len24_executor_target_mass": 0.834678590297699,
          "len24_init_accuracy": 1.0,
          "len24_op_accuracy": 0.9943033854166666,
          "len24_arg_accuracy": 1.0,
          "len24_program_exact": 0.87109375
        },
        {
          "stage": "answer_retention",
          "local_step": 1,
          "step": 801,
          "loss": 0.05444745719432831,
          "trace_loss_active": false,
          "lr_scale": 1.0,
          "executor_loss": 0.05444745719432831,
          "len4_direct_accuracy": NaN,
          "len4_direct_target_mass": NaN,
          "len4_executor_accuracy": 0.97265625,
          "len4_executor_target_mass": 0.9723573923110962,
          "len4_init_accuracy": 1.0,
          "len4_op_accuracy": 0.9931640625,
          "len4_arg_accuracy": 1.0,
          "len4_program_exact": 0.97265625,
          "len8_direct_accuracy": NaN,
          "len8_direct_target_mass": NaN,
          "len8_executor_accuracy": 0.9609375,
          "len8_executor_target_mass": 0.9583216905593872,
          "len8_init_accuracy": 1.0,
          "len8_op_accuracy": 0.9951171875,
          "len8_arg_accuracy": 1.0,
          "len8_program_exact": 0.9609375,
          "len12_direct_accuracy": NaN,
          "len12_direct_target_mass": NaN,
          "len12_executor_accuracy": 0.9140625,
          "len12_executor_target_mass": 0.8962996006011963,
          "len12_init_accuracy": 1.0,
          "len12_op_accuracy": 0.9925130208333334,
          "len12_arg_accuracy": 1.0,
          "len12_program_exact": 0.9140625,
          "len24_direct_accuracy": NaN,
          "len24_direct_target_mass": NaN,
          "len24_executor_accuracy": 0.75390625,
          "len24_executor_target_mass": 0.7152934074401855,
          "len24_init_accuracy": 1.0,
          "len24_op_accuracy": 0.9895833333333334,
          "len24_arg_accuracy": 1.0,
          "len24_program_exact": 0.75390625
        },
        {
          "stage": "answer_retention",
          "local_step": 400,
          "step": 1200,
          "loss": 0.00010546730482019484,
          "trace_loss_active": false,
          "lr_scale": 1.0,
          "executor_loss": 0.00010546730482019484,
          "len4_direct_accuracy": NaN,
          "len4_direct_target_mass": NaN,
          "len4_executor_accuracy": 1.0,
          "len4_executor_target_mass": 0.9995648860931396,
          "len4_init_accuracy": 1.0,
          "len4_op_accuracy": 1.0,
          "len4_arg_accuracy": 1.0,
          "len4_program_exact": 1.0,
          "len8_direct_accuracy": NaN,
          "len8_direct_target_mass": NaN,
          "len8_executor_accuracy": 1.0,
          "len8_executor_target_mass": 0.999672532081604,
          "len8_init_accuracy": 1.0,
          "len8_op_accuracy": 1.0,
          "len8_arg_accuracy": 1.0,
          "len8_program_exact": 1.0,
          "len12_direct_accuracy": NaN,
          "len12_direct_target_mass": NaN,
          "len12_executor_accuracy": 0.99609375,
          "len12_executor_target_mass": 0.9958053827285767,
          "len12_init_accuracy": 1.0,
          "len12_op_accuracy": 0.9996744791666666,
          "len12_arg_accuracy": 1.0,
          "len12_program_exact": 0.99609375,
          "len24_direct_accuracy": NaN,
          "len24_direct_target_mass": NaN,
          "len24_executor_accuracy": 0.96875,
          "len24_executor_target_mass": 0.9523723125457764,
          "len24_init_accuracy": 1.0,
          "len24_op_accuracy": 0.9986979166666666,
          "len24_arg_accuracy": 1.0,
          "len24_program_exact": 0.96875
        },
        {
          "stage": "answer_retention",
          "local_step": 800,
          "step": 1600,
          "loss": 6.374481745297089e-05,
          "trace_loss_active": false,
          "lr_scale": 1.0,
          "executor_loss": 6.374481745297089e-05,
          "len4_direct_accuracy": NaN,
          "len4_direct_target_mass": NaN,
          "len4_executor_accuracy": 1.0,
          "len4_executor_target_mass": 0.9996262788772583,
          "len4_init_accuracy": 1.0,
          "len4_op_accuracy": 1.0,
          "len4_arg_accuracy": 1.0,
          "len4_program_exact": 1.0,
          "len8_direct_accuracy": NaN,
          "len8_direct_target_mass": NaN,
          "len8_executor_accuracy": 1.0,
          "len8_executor_target_mass": 0.9997516870498657,
          "len8_init_accuracy": 1.0,
          "len8_op_accuracy": 1.0,
          "len8_arg_accuracy": 1.0,
          "len8_program_exact": 1.0,
          "len12_direct_accuracy": NaN,
          "len12_direct_target_mass": NaN,
          "len12_executor_accuracy": 0.99609375,
          "len12_executor_target_mass": 0.9954250454902649,
          "len12_init_accuracy": 1.0,
          "len12_op_accuracy": 0.9996744791666666,
          "len12_arg_accuracy": 1.0,
          "len12_program_exact": 0.99609375,
          "len24_direct_accuracy": NaN,
          "len24_direct_target_mass": NaN,
          "len24_executor_accuracy": 0.96875,
          "len24_executor_target_mass": 0.9532598853111267,
          "len24_init_accuracy": 1.0,
          "len24_op_accuracy": 0.9986979166666666,
          "len24_arg_accuracy": 1.0,
          "len24_program_exact": 0.96875
        }
      ],
      "final_metrics": {
        "len4": {
          "variant": "compiler_trace_then_answer",
          "n": 256.0,
          "direct_accuracy": NaN,
          "direct_target_mass": NaN,
          "executor_accuracy": 1.0,
          "executor_target_mass": 0.9996262788772583,
          "init_accuracy": 1.0,
          "op_accuracy": 1.0,
          "arg_accuracy": 1.0,
          "program_exact": 1.0
        },
        "len8": {
          "variant": "compiler_trace_then_answer",
          "n": 256.0,
          "direct_accuracy": NaN,
          "direct_target_mass": NaN,
          "executor_accuracy": 1.0,
          "executor_target_mass": 0.9997516870498657,
          "init_accuracy": 1.0,
          "op_accuracy": 1.0,
          "arg_accuracy": 1.0,
          "program_exact": 1.0
        },
        "len12": {
          "variant": "compiler_trace_then_answer",
          "n": 256.0,
          "direct_accuracy": NaN,
          "direct_target_mass": NaN,
          "executor_accuracy": 0.99609375,
          "executor_target_mass": 0.9954250454902649,
          "init_accuracy": 1.0,
          "op_accuracy": 0.9996744791666666,
          "arg_accuracy": 1.0,
          "program_exact": 0.99609375
        },
        "len24": {
          "variant": "compiler_trace_then_answer",
          "n": 256.0,
          "direct_accuracy": NaN,
          "direct_target_mass": NaN,
          "executor_accuracy": 0.96875,
          "executor_target_mass": 0.9532598853111267,
          "init_accuracy": 1.0,
          "op_accuracy": 0.9986979166666666,
          "arg_accuracy": 1.0,
          "program_exact": 0.96875
        }
      }
    },
    "compiler_trace_then_answer_low_lr": {
      "variant": "compiler_trace_then_answer_low_lr",
      "train_seconds": 103.92612409591675,
      "checkpoints": [
        "large_artifacts/qwen_trace_bootstrap_retention/checkpoints/main_qwen35_retention/compiler_trace_then_answer_low_lr/checkpoint_final.pt"
      ],
      "train_log": [
        {
          "stage": "trace_bootstrap",
          "local_step": 1,
          "step": 1,
          "loss": 14.824259757995605,
          "trace_loss_active": true,
          "lr_scale": 1.0,
          "executor_loss": 4.579751014709473,
          "init_loss": 4.582459449768066,
          "op_loss": 1.09193754196167,
          "arg_loss": 4.570111274719238,
          "len4_direct_accuracy": NaN,
          "len4_direct_target_mass": NaN,
          "len4_executor_accuracy": 0.015625,
          "len4_executor_target_mass": 0.010310400277376175,
          "len4_init_accuracy": 0.02734375,
          "len4_op_accuracy": 0.310546875,
          "len4_arg_accuracy": 0.185546875,
          "len4_program_exact": 0.0,
          "len8_direct_accuracy": NaN,
          "len8_direct_target_mass": NaN,
          "len8_executor_accuracy": 0.0078125,
          "len8_executor_target_mass": 0.01030952762812376,
          "len8_init_accuracy": 0.03125,
          "len8_op_accuracy": 0.32861328125,
          "len8_arg_accuracy": 0.162109375,
          "len8_program_exact": 0.0,
          "len12_direct_accuracy": NaN,
          "len12_direct_target_mass": NaN,
          "len12_executor_accuracy": 0.0078125,
          "len12_executor_target_mass": 0.010309104807674885,
          "len12_init_accuracy": 0.0546875,
          "len12_op_accuracy": 0.3349609375,
          "len12_arg_accuracy": 0.1806640625,
          "len12_program_exact": 0.0,
          "len24_direct_accuracy": NaN,
          "len24_direct_target_mass": NaN,
          "len24_executor_accuracy": 0.015625,
          "len24_executor_target_mass": 0.01030734647065401,
          "len24_init_accuracy": 0.03515625,
          "len24_op_accuracy": 0.3317057291666667,
          "len24_arg_accuracy": 0.16715494791666666,
          "len24_program_exact": 0.0
        },
        {
          "stage": "trace_bootstrap",
          "local_step": 400,
          "step": 400,
          "loss": 0.0007649306789971888,
          "trace_loss_active": true,
          "lr_scale": 1.0,
          "executor_loss": 0.00042066589230671525,
          "init_loss": 0.00028347905026748776,
          "op_loss": 4.429532509675482e-06,
          "arg_loss": 5.635617708321661e-05,
          "len4_direct_accuracy": NaN,
          "len4_direct_target_mass": NaN,
          "len4_executor_accuracy": 1.0,
          "len4_executor_target_mass": 0.9976519346237183,
          "len4_init_accuracy": 1.0,
          "len4_op_accuracy": 1.0,
          "len4_arg_accuracy": 1.0,
          "len4_program_exact": 1.0,
          "len8_direct_accuracy": NaN,
          "len8_direct_target_mass": NaN,
          "len8_executor_accuracy": 0.98828125,
          "len8_executor_target_mass": 0.9837839603424072,
          "len8_init_accuracy": 1.0,
          "len8_op_accuracy": 0.99853515625,
          "len8_arg_accuracy": 1.0,
          "len8_program_exact": 0.98828125,
          "len12_direct_accuracy": NaN,
          "len12_direct_target_mass": NaN,
          "len12_executor_accuracy": 0.9375,
          "len12_executor_target_mass": 0.9317259192466736,
          "len12_init_accuracy": 1.0,
          "len12_op_accuracy": 0.9944661458333334,
          "len12_arg_accuracy": 1.0,
          "len12_program_exact": 0.9375,
          "len24_direct_accuracy": NaN,
          "len24_direct_target_mass": NaN,
          "len24_executor_accuracy": 0.87109375,
          "len24_executor_target_mass": 0.8214964270591736,
          "len24_init_accuracy": 1.0,
          "len24_op_accuracy": 0.9939778645833334,
          "len24_arg_accuracy": 1.0,
          "len24_program_exact": 0.87109375
        },
        {
          "stage": "trace_bootstrap",
          "local_step": 800,
          "step": 800,
          "loss": 0.00023410184076055884,
          "trace_loss_active": true,
          "lr_scale": 1.0,
          "executor_loss": 0.00012916112609673291,
          "init_loss": 8.826862904243171e-05,
          "op_loss": 1.0299484074494103e-06,
          "arg_loss": 1.56421465362655e-05,
          "len4_direct_accuracy": NaN,
          "len4_direct_target_mass": NaN,
          "len4_executor_accuracy": 1.0,
          "len4_executor_target_mass": 0.9981541633605957,
          "len4_init_accuracy": 1.0,
          "len4_op_accuracy": 1.0,
          "len4_arg_accuracy": 1.0,
          "len4_program_exact": 1.0,
          "len8_direct_accuracy": NaN,
          "len8_direct_target_mass": NaN,
          "len8_executor_accuracy": 0.98828125,
          "len8_executor_target_mass": 0.9865819215774536,
          "len8_init_accuracy": 1.0,
          "len8_op_accuracy": 0.99853515625,
          "len8_arg_accuracy": 1.0,
          "len8_program_exact": 0.98828125,
          "len12_direct_accuracy": NaN,
          "len12_direct_target_mass": NaN,
          "len12_executor_accuracy": 0.9375,
          "len12_executor_target_mass": 0.9375402927398682,
          "len12_init_accuracy": 1.0,
          "len12_op_accuracy": 0.9947916666666666,
          "len12_arg_accuracy": 1.0,
          "len12_program_exact": 0.9375,
          "len24_direct_accuracy": NaN,
          "len24_direct_target_mass": NaN,
          "len24_executor_accuracy": 0.87109375,
          "len24_executor_target_mass": 0.834678590297699,
          "len24_init_accuracy": 1.0,
          "len24_op_accuracy": 0.9943033854166666,
          "len24_arg_accuracy": 1.0,
          "len24_program_exact": 0.87109375
        },
        {
          "stage": "answer_retention_low_lr",
          "local_step": 1,
          "step": 801,
          "loss": 0.05444745719432831,
          "trace_loss_active": false,
          "lr_scale": 0.1,
          "executor_loss": 0.05444745719432831,
          "len4_direct_accuracy": NaN,
          "len4_direct_target_mass": NaN,
          "len4_executor_accuracy": 1.0,
          "len4_executor_target_mass": 0.9990960359573364,
          "len4_init_accuracy": 1.0,
          "len4_op_accuracy": 1.0,
          "len4_arg_accuracy": 1.0,
          "len4_program_exact": 1.0,
          "len8_direct_accuracy": NaN,
          "len8_direct_target_mass": NaN,
          "len8_executor_accuracy": 0.98828125,
          "len8_executor_target_mass": 0.9887381792068481,
          "len8_init_accuracy": 1.0,
          "len8_op_accuracy": 0.99853515625,
          "len8_arg_accuracy": 1.0,
          "len8_program_exact": 0.98828125,
          "len12_direct_accuracy": NaN,
          "len12_direct_target_mass": NaN,
          "len12_executor_accuracy": 0.94921875,
          "len12_executor_target_mass": 0.9443823099136353,
          "len12_init_accuracy": 1.0,
          "len12_op_accuracy": 0.9957682291666666,
          "len12_arg_accuracy": 1.0,
          "len12_program_exact": 0.94921875,
          "len24_direct_accuracy": NaN,
          "len24_direct_target_mass": NaN,
          "len24_executor_accuracy": 0.8984375,
          "len24_executor_target_mass": 0.8545814752578735,
          "len24_init_accuracy": 1.0,
          "len24_op_accuracy": 0.9957682291666666,
          "len24_arg_accuracy": 1.0,
          "len24_program_exact": 0.8984375
        },
        {
          "stage": "answer_retention_low_lr",
          "local_step": 400,
          "step": 1200,
          "loss": 0.00020508143643382937,
          "trace_loss_active": false,
          "lr_scale": 0.1,
          "executor_loss": 0.00020508143643382937,
          "len4_direct_accuracy": NaN,
          "len4_direct_target_mass": NaN,
          "len4_executor_accuracy": 1.0,
          "len4_executor_target_mass": 0.9994431138038635,
          "len4_init_accuracy": 1.0,
          "len4_op_accuracy": 1.0,
          "len4_arg_accuracy": 1.0,
          "len4_program_exact": 1.0,
          "len8_direct_accuracy": NaN,
          "len8_direct_target_mass": NaN,
          "len8_executor_accuracy": 1.0,
          "len8_executor_target_mass": 0.9974724054336548,
          "len8_init_accuracy": 1.0,
          "len8_op_accuracy": 1.0,
          "len8_arg_accuracy": 1.0,
          "len8_program_exact": 1.0,
          "len12_direct_accuracy": NaN,
          "len12_direct_target_mass": NaN,
          "len12_executor_accuracy": 0.98828125,
          "len12_executor_target_mass": 0.9842265844345093,
          "len12_init_accuracy": 1.0,
          "len12_op_accuracy": 0.9990234375,
          "len12_arg_accuracy": 1.0,
          "len12_program_exact": 0.98828125,
          "len24_direct_accuracy": NaN,
          "len24_direct_target_mass": NaN,
          "len24_executor_accuracy": 0.953125,
          "len24_executor_target_mass": 0.9184905290603638,
          "len24_init_accuracy": 1.0,
          "len24_op_accuracy": 0.998046875,
          "len24_arg_accuracy": 1.0,
          "len24_program_exact": 0.953125
        },
        {
          "stage": "answer_retention_low_lr",
          "local_step": 800,
          "step": 1600,
          "loss": 0.00016034470172598958,
          "trace_loss_active": false,
          "lr_scale": 0.1,
          "executor_loss": 0.00016034470172598958,
          "len4_direct_accuracy": NaN,
          "len4_direct_target_mass": NaN,
          "len4_executor_accuracy": 1.0,
          "len4_executor_target_mass": 0.9994856119155884,
          "len4_init_accuracy": 1.0,
          "len4_op_accuracy": 1.0,
          "len4_arg_accuracy": 1.0,
          "len4_program_exact": 1.0,
          "len8_direct_accuracy": NaN,
          "len8_direct_target_mass": NaN,
          "len8_executor_accuracy": 1.0,
          "len8_executor_target_mass": 0.9977781176567078,
          "len8_init_accuracy": 1.0,
          "len8_op_accuracy": 1.0,
          "len8_arg_accuracy": 1.0,
          "len8_program_exact": 1.0,
          "len12_direct_accuracy": NaN,
          "len12_direct_target_mass": NaN,
          "len12_executor_accuracy": 0.9921875,
          "len12_executor_target_mass": 0.9858863949775696,
          "len12_init_accuracy": 1.0,
          "len12_op_accuracy": 0.9993489583333334,
          "len12_arg_accuracy": 1.0,
          "len12_program_exact": 0.9921875,
          "len24_direct_accuracy": NaN,
          "len24_direct_target_mass": NaN,
          "len24_executor_accuracy": 0.95703125,
          "len24_executor_target_mass": 0.9284825921058655,
          "len24_init_accuracy": 1.0,
          "len24_op_accuracy": 0.9982096354166666,
          "len24_arg_accuracy": 1.0,
          "len24_program_exact": 0.95703125
        }
      ],
      "final_metrics": {
        "len4": {
          "variant": "compiler_trace_then_answer_low_lr",
          "n": 256.0,
          "direct_accuracy": NaN,
          "direct_target_mass": NaN,
          "executor_accuracy": 1.0,
          "executor_target_mass": 0.9994856119155884,
          "init_accuracy": 1.0,
          "op_accuracy": 1.0,
          "arg_accuracy": 1.0,
          "program_exact": 1.0
        },
        "len8": {
          "variant": "compiler_trace_then_answer_low_lr",
          "n": 256.0,
          "direct_accuracy": NaN,
          "direct_target_mass": NaN,
          "executor_accuracy": 1.0,
          "executor_target_mass": 0.9977781176567078,
          "init_accuracy": 1.0,
          "op_accuracy": 1.0,
          "arg_accuracy": 1.0,
          "program_exact": 1.0
        },
        "len12": {
          "variant": "compiler_trace_then_answer_low_lr",
          "n": 256.0,
          "direct_accuracy": NaN,
          "direct_target_mass": NaN,
          "executor_accuracy": 0.9921875,
          "executor_target_mass": 0.9858863949775696,
          "init_accuracy": 1.0,
          "op_accuracy": 0.9993489583333334,
          "arg_accuracy": 1.0,
          "program_exact": 0.9921875
        },
        "len24": {
          "variant": "compiler_trace_then_answer_low_lr",
          "n": 256.0,
          "direct_accuracy": NaN,
          "direct_target_mass": NaN,
          "executor_accuracy": 0.95703125,
          "executor_target_mass": 0.9284825921058655,
          "init_accuracy": 1.0,
          "op_accuracy": 0.9982096354166666,
          "arg_accuracy": 1.0,
          "program_exact": 0.95703125
        }
      }
    }
  }
}