{
  "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": 12,
    "train_min_len": 1,
    "train_max_len": 3,
    "answer_train_min_len": 1,
    "answer_train_max_len": 6,
    "eval_lengths": "3,6,12",
    "train_size": 192,
    "answer_train_size": 192,
    "eval_size": 64,
    "feature_batch_size": 4,
    "train_batch_size": 64,
    "eval_batch_size": 64,
    "bootstrap_steps": 160,
    "answer_steps": 160,
    "eval_every": 80,
    "stage_eval_every": 80,
    "lr": 0.001,
    "weight_decay": 0.01,
    "grad_clip": 1.0,
    "head_width": 384,
    "trace_loss_weight": 1.0,
    "executor_loss_weight": 1.0,
    "variants": "direct,compiler_trace,compiler_answer_only,compiler_trace_then_answer",
    "max_length": 512,
    "seed": 7,
    "output_dir": "experiments/qwen_trace_bootstrap_retention/runs/pilot_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": 192,
    "answer_train_size": 192,
    "eval_size": 64,
    "bootstrap_train_lengths": [
      1,
      3
    ],
    "answer_train_lengths": [
      1,
      6
    ],
    "eval_lengths": [
      3,
      6,
      12
    ]
  },
  "variants": {
    "direct": {
      "variant": "direct",
      "train_seconds": 7.750715255737305,
      "checkpoints": [
        "large_artifacts/qwen_trace_bootstrap_retention/checkpoints/pilot_qwen35_retention/direct/checkpoint_final.pt"
      ],
      "train_log": [
        {
          "stage": "answer_only",
          "local_step": 1,
          "step": 1,
          "loss": 4.625516891479492,
          "trace_loss_active": false,
          "direct_loss": 4.625516891479492,
          "len3_direct_accuracy": 0.0,
          "len3_direct_target_mass": 0.01249202061444521,
          "len3_executor_accuracy": NaN,
          "len3_executor_target_mass": NaN,
          "len3_init_accuracy": NaN,
          "len3_op_accuracy": NaN,
          "len3_arg_accuracy": NaN,
          "len3_program_exact": NaN,
          "len6_direct_accuracy": 0.015625,
          "len6_direct_target_mass": 0.012065712362527847,
          "len6_executor_accuracy": NaN,
          "len6_executor_target_mass": NaN,
          "len6_init_accuracy": NaN,
          "len6_op_accuracy": NaN,
          "len6_arg_accuracy": NaN,
          "len6_program_exact": NaN,
          "len12_direct_accuracy": 0.0,
          "len12_direct_target_mass": 0.010141589678823948,
          "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
        },
        {
          "stage": "answer_only",
          "local_step": 80,
          "step": 80,
          "loss": 3.870643138885498,
          "trace_loss_active": false,
          "direct_loss": 3.870643138885498,
          "len3_direct_accuracy": 0.0,
          "len3_direct_target_mass": 0.011254181154072285,
          "len3_executor_accuracy": NaN,
          "len3_executor_target_mass": NaN,
          "len3_init_accuracy": NaN,
          "len3_op_accuracy": NaN,
          "len3_arg_accuracy": NaN,
          "len3_program_exact": NaN,
          "len6_direct_accuracy": 0.0,
          "len6_direct_target_mass": 0.006705072708427906,
          "len6_executor_accuracy": NaN,
          "len6_executor_target_mass": NaN,
          "len6_init_accuracy": NaN,
          "len6_op_accuracy": NaN,
          "len6_arg_accuracy": NaN,
          "len6_program_exact": NaN,
          "len12_direct_accuracy": 0.0,
          "len12_direct_target_mass": 0.008576484397053719,
          "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
        },
        {
          "stage": "answer_only",
          "local_step": 160,
          "step": 160,
          "loss": 3.426636219024658,
          "trace_loss_active": false,
          "direct_loss": 3.426636219024658,
          "len3_direct_accuracy": 0.0,
          "len3_direct_target_mass": 0.01028795912861824,
          "len3_executor_accuracy": NaN,
          "len3_executor_target_mass": NaN,
          "len3_init_accuracy": NaN,
          "len3_op_accuracy": NaN,
          "len3_arg_accuracy": NaN,
          "len3_program_exact": NaN,
          "len6_direct_accuracy": 0.0,
          "len6_direct_target_mass": 0.008024861104786396,
          "len6_executor_accuracy": NaN,
          "len6_executor_target_mass": NaN,
          "len6_init_accuracy": NaN,
          "len6_op_accuracy": NaN,
          "len6_arg_accuracy": NaN,
          "len6_program_exact": NaN,
          "len12_direct_accuracy": 0.0,
          "len12_direct_target_mass": 0.010308399796485901,
          "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
        },
        {
          "stage": "answer_only",
          "local_step": 240,
          "step": 240,
          "loss": 3.0883116722106934,
          "trace_loss_active": false,
          "direct_loss": 3.0883116722106934,
          "len3_direct_accuracy": 0.0,
          "len3_direct_target_mass": 0.012687250971794128,
          "len3_executor_accuracy": NaN,
          "len3_executor_target_mass": NaN,
          "len3_init_accuracy": NaN,
          "len3_op_accuracy": NaN,
          "len3_arg_accuracy": NaN,
          "len3_program_exact": NaN,
          "len6_direct_accuracy": 0.0,
          "len6_direct_target_mass": 0.006337698083370924,
          "len6_executor_accuracy": NaN,
          "len6_executor_target_mass": NaN,
          "len6_init_accuracy": NaN,
          "len6_op_accuracy": NaN,
          "len6_arg_accuracy": NaN,
          "len6_program_exact": NaN,
          "len12_direct_accuracy": 0.0,
          "len12_direct_target_mass": 0.007270154543220997,
          "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
        },
        {
          "stage": "answer_only",
          "local_step": 320,
          "step": 320,
          "loss": 2.609973907470703,
          "trace_loss_active": false,
          "direct_loss": 2.609973907470703,
          "len3_direct_accuracy": 0.015625,
          "len3_direct_target_mass": 0.010394666343927383,
          "len3_executor_accuracy": NaN,
          "len3_executor_target_mass": NaN,
          "len3_init_accuracy": NaN,
          "len3_op_accuracy": NaN,
          "len3_arg_accuracy": NaN,
          "len3_program_exact": NaN,
          "len6_direct_accuracy": 0.0,
          "len6_direct_target_mass": 0.005934556946158409,
          "len6_executor_accuracy": NaN,
          "len6_executor_target_mass": NaN,
          "len6_init_accuracy": NaN,
          "len6_op_accuracy": NaN,
          "len6_arg_accuracy": NaN,
          "len6_program_exact": NaN,
          "len12_direct_accuracy": 0.015625,
          "len12_direct_target_mass": 0.012585785239934921,
          "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
        }
      ],
      "final_metrics": {
        "len3": {
          "variant": "direct",
          "n": 64.0,
          "direct_accuracy": 0.015625,
          "direct_target_mass": 0.010394666343927383,
          "executor_accuracy": NaN,
          "executor_target_mass": NaN,
          "init_accuracy": NaN,
          "op_accuracy": NaN,
          "arg_accuracy": NaN,
          "program_exact": NaN
        },
        "len6": {
          "variant": "direct",
          "n": 64.0,
          "direct_accuracy": 0.0,
          "direct_target_mass": 0.005934556946158409,
          "executor_accuracy": NaN,
          "executor_target_mass": NaN,
          "init_accuracy": NaN,
          "op_accuracy": NaN,
          "arg_accuracy": NaN,
          "program_exact": NaN
        },
        "len12": {
          "variant": "direct",
          "n": 64.0,
          "direct_accuracy": 0.015625,
          "direct_target_mass": 0.012585785239934921,
          "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": 15.179947137832642,
      "checkpoints": [
        "large_artifacts/qwen_trace_bootstrap_retention/checkpoints/pilot_qwen35_retention/compiler_trace/checkpoint_final.pt"
      ],
      "train_log": [
        {
          "stage": "trace_bootstrap",
          "local_step": 1,
          "step": 1,
          "loss": 14.926406860351562,
          "trace_loss_active": true,
          "executor_loss": 4.579817295074463,
          "init_loss": 4.567264556884766,
          "op_loss": 1.1324918270111084,
          "arg_loss": 4.646833896636963,
          "len3_direct_accuracy": NaN,
          "len3_direct_target_mass": NaN,
          "len3_executor_accuracy": 0.03125,
          "len3_executor_target_mass": 0.01027483306825161,
          "len3_init_accuracy": 0.0,
          "len3_op_accuracy": 0.5989583333333334,
          "len3_arg_accuracy": 0.057291666666666664,
          "len3_program_exact": 0.0,
          "len6_direct_accuracy": NaN,
          "len6_direct_target_mass": NaN,
          "len6_executor_accuracy": 0.03125,
          "len6_executor_target_mass": 0.010326087474822998,
          "len6_init_accuracy": 0.0,
          "len6_op_accuracy": 0.4973958333333333,
          "len6_arg_accuracy": 0.06510416666666667,
          "len6_program_exact": 0.0,
          "len12_direct_accuracy": NaN,
          "len12_direct_target_mass": NaN,
          "len12_executor_accuracy": 0.015625,
          "len12_executor_target_mass": 0.010244632139801979,
          "len12_init_accuracy": 0.0,
          "len12_op_accuracy": 0.4973958333333333,
          "len12_arg_accuracy": 0.059895833333333336,
          "len12_program_exact": 0.0
        },
        {
          "stage": "trace_bootstrap",
          "local_step": 80,
          "step": 80,
          "loss": 0.3942287266254425,
          "trace_loss_active": true,
          "executor_loss": 0.20208758115768433,
          "init_loss": 0.18128947913646698,
          "op_loss": 0.0004542924289125949,
          "arg_loss": 0.010397372767329216,
          "len3_direct_accuracy": NaN,
          "len3_direct_target_mass": NaN,
          "len3_executor_accuracy": 0.75,
          "len3_executor_target_mass": 0.5053077936172485,
          "len3_init_accuracy": 0.796875,
          "len3_op_accuracy": 1.0,
          "len3_arg_accuracy": 0.984375,
          "len3_program_exact": 0.75,
          "len6_direct_accuracy": NaN,
          "len6_direct_target_mass": NaN,
          "len6_executor_accuracy": 0.765625,
          "len6_executor_target_mass": 0.4526320695877075,
          "len6_init_accuracy": 0.90625,
          "len6_op_accuracy": 0.9817708333333334,
          "len6_arg_accuracy": 0.9947916666666666,
          "len6_program_exact": 0.765625,
          "len12_direct_accuracy": NaN,
          "len12_direct_target_mass": NaN,
          "len12_executor_accuracy": 0.4375,
          "len12_executor_target_mass": 0.17830955982208252,
          "len12_init_accuracy": 0.796875,
          "len12_op_accuracy": 0.9518229166666666,
          "len12_arg_accuracy": 0.9973958333333334,
          "len12_program_exact": 0.4375
        },
        {
          "stage": "trace_bootstrap",
          "local_step": 160,
          "step": 160,
          "loss": 0.007986078038811684,
          "trace_loss_active": true,
          "executor_loss": 0.004319241736084223,
          "init_loss": 0.003024287987500429,
          "op_loss": 3.165808448102325e-05,
          "arg_loss": 0.0006108905072323978,
          "len3_direct_accuracy": NaN,
          "len3_direct_target_mass": NaN,
          "len3_executor_accuracy": 0.828125,
          "len3_executor_target_mass": 0.7882490158081055,
          "len3_init_accuracy": 0.828125,
          "len3_op_accuracy": 1.0,
          "len3_arg_accuracy": 1.0,
          "len3_program_exact": 0.828125,
          "len6_direct_accuracy": NaN,
          "len6_direct_target_mass": NaN,
          "len6_executor_accuracy": 0.796875,
          "len6_executor_target_mass": 0.7468955516815186,
          "len6_init_accuracy": 0.9375,
          "len6_op_accuracy": 0.9765625,
          "len6_arg_accuracy": 1.0,
          "len6_program_exact": 0.796875,
          "len12_direct_accuracy": NaN,
          "len12_direct_target_mass": NaN,
          "len12_executor_accuracy": 0.46875,
          "len12_executor_target_mass": 0.35988953709602356,
          "len12_init_accuracy": 0.890625,
          "len12_op_accuracy": 0.9375,
          "len12_arg_accuracy": 1.0,
          "len12_program_exact": 0.46875
        },
        {
          "stage": "trace_continuation",
          "local_step": 1,
          "step": 161,
          "loss": 5.902594566345215,
          "trace_loss_active": true,
          "executor_loss": 1.7929182052612305,
          "init_loss": 4.064006805419922,
          "op_loss": 0.04075169190764427,
          "arg_loss": 0.004918315447866917,
          "len3_direct_accuracy": NaN,
          "len3_direct_target_mass": NaN,
          "len3_executor_accuracy": 0.8125,
          "len3_executor_target_mass": 0.784636378288269,
          "len3_init_accuracy": 0.828125,
          "len3_op_accuracy": 1.0,
          "len3_arg_accuracy": 0.9947916666666666,
          "len3_program_exact": 0.8125,
          "len6_direct_accuracy": NaN,
          "len6_direct_target_mass": NaN,
          "len6_executor_accuracy": 0.8125,
          "len6_executor_target_mass": 0.7370574474334717,
          "len6_init_accuracy": 0.9375,
          "len6_op_accuracy": 0.9791666666666666,
          "len6_arg_accuracy": 1.0,
          "len6_program_exact": 0.8125,
          "len12_direct_accuracy": NaN,
          "len12_direct_target_mass": NaN,
          "len12_executor_accuracy": 0.578125,
          "len12_executor_target_mass": 0.4098951816558838,
          "len12_init_accuracy": 0.890625,
          "len12_op_accuracy": 0.96875,
          "len12_arg_accuracy": 0.9973958333333334,
          "len12_program_exact": 0.578125
        },
        {
          "stage": "trace_continuation",
          "local_step": 80,
          "step": 240,
          "loss": 0.004094962030649185,
          "trace_loss_active": true,
          "executor_loss": 0.0024617905728518963,
          "init_loss": 0.0012996225850656629,
          "op_loss": 0.00013459300680551678,
          "arg_loss": 0.00019895558943971992,
          "len3_direct_accuracy": NaN,
          "len3_direct_target_mass": NaN,
          "len3_executor_accuracy": 1.0,
          "len3_executor_target_mass": 0.9795650243759155,
          "len3_init_accuracy": 1.0,
          "len3_op_accuracy": 1.0,
          "len3_arg_accuracy": 1.0,
          "len3_program_exact": 1.0,
          "len6_direct_accuracy": NaN,
          "len6_direct_target_mass": NaN,
          "len6_executor_accuracy": 0.984375,
          "len6_executor_target_mass": 0.9518207311630249,
          "len6_init_accuracy": 0.984375,
          "len6_op_accuracy": 1.0,
          "len6_arg_accuracy": 1.0,
          "len6_program_exact": 0.984375,
          "len12_direct_accuracy": NaN,
          "len12_direct_target_mass": NaN,
          "len12_executor_accuracy": 0.9375,
          "len12_executor_target_mass": 0.8843073844909668,
          "len12_init_accuracy": 0.96875,
          "len12_op_accuracy": 0.9973958333333334,
          "len12_arg_accuracy": 1.0,
          "len12_program_exact": 0.9375
        },
        {
          "stage": "trace_continuation",
          "local_step": 160,
          "step": 320,
          "loss": 0.0019455056171864271,
          "trace_loss_active": true,
          "executor_loss": 0.001081803347915411,
          "init_loss": 0.0007734072278253734,
          "op_loss": 2.616523033793783e-06,
          "arg_loss": 8.767851977609098e-05,
          "len3_direct_accuracy": NaN,
          "len3_direct_target_mass": NaN,
          "len3_executor_accuracy": 1.0,
          "len3_executor_target_mass": 0.9817734360694885,
          "len3_init_accuracy": 1.0,
          "len3_op_accuracy": 1.0,
          "len3_arg_accuracy": 1.0,
          "len3_program_exact": 1.0,
          "len6_direct_accuracy": NaN,
          "len6_direct_target_mass": NaN,
          "len6_executor_accuracy": 0.984375,
          "len6_executor_target_mass": 0.9615117311477661,
          "len6_init_accuracy": 0.984375,
          "len6_op_accuracy": 1.0,
          "len6_arg_accuracy": 1.0,
          "len6_program_exact": 0.984375,
          "len12_direct_accuracy": NaN,
          "len12_direct_target_mass": NaN,
          "len12_executor_accuracy": 0.921875,
          "len12_executor_target_mass": 0.9132857322692871,
          "len12_init_accuracy": 0.9375,
          "len12_op_accuracy": 0.9973958333333334,
          "len12_arg_accuracy": 1.0,
          "len12_program_exact": 0.921875
        }
      ],
      "final_metrics": {
        "len3": {
          "variant": "compiler_trace",
          "n": 64.0,
          "direct_accuracy": NaN,
          "direct_target_mass": NaN,
          "executor_accuracy": 1.0,
          "executor_target_mass": 0.9817734360694885,
          "init_accuracy": 1.0,
          "op_accuracy": 1.0,
          "arg_accuracy": 1.0,
          "program_exact": 1.0
        },
        "len6": {
          "variant": "compiler_trace",
          "n": 64.0,
          "direct_accuracy": NaN,
          "direct_target_mass": NaN,
          "executor_accuracy": 0.984375,
          "executor_target_mass": 0.9615117311477661,
          "init_accuracy": 0.984375,
          "op_accuracy": 1.0,
          "arg_accuracy": 1.0,
          "program_exact": 0.984375
        },
        "len12": {
          "variant": "compiler_trace",
          "n": 64.0,
          "direct_accuracy": NaN,
          "direct_target_mass": NaN,
          "executor_accuracy": 0.921875,
          "executor_target_mass": 0.9132857322692871,
          "init_accuracy": 0.9375,
          "op_accuracy": 0.9973958333333334,
          "arg_accuracy": 1.0,
          "program_exact": 0.921875
        }
      }
    },
    "compiler_answer_only": {
      "variant": "compiler_answer_only",
      "train_seconds": 14.591087102890015,
      "checkpoints": [
        "large_artifacts/qwen_trace_bootstrap_retention/checkpoints/pilot_qwen35_retention/compiler_answer_only/checkpoint_final.pt"
      ],
      "train_log": [
        {
          "stage": "answer_only",
          "local_step": 1,
          "step": 1,
          "loss": 4.579855918884277,
          "trace_loss_active": false,
          "executor_loss": 4.579855918884277,
          "len3_direct_accuracy": NaN,
          "len3_direct_target_mass": NaN,
          "len3_executor_accuracy": 0.015625,
          "len3_executor_target_mass": 0.010309549048542976,
          "len3_init_accuracy": 0.015625,
          "len3_op_accuracy": 0.3125,
          "len3_arg_accuracy": 0.0,
          "len3_program_exact": 0.0,
          "len6_direct_accuracy": NaN,
          "len6_direct_target_mass": NaN,
          "len6_executor_accuracy": 0.0,
          "len6_executor_target_mass": 0.010309278033673763,
          "len6_init_accuracy": 0.0,
          "len6_op_accuracy": 0.2994791666666667,
          "len6_arg_accuracy": 0.0,
          "len6_program_exact": 0.0,
          "len12_direct_accuracy": NaN,
          "len12_direct_target_mass": NaN,
          "len12_executor_accuracy": 0.015625,
          "len12_executor_target_mass": 0.010309278033673763,
          "len12_init_accuracy": 0.015625,
          "len12_op_accuracy": 0.3515625,
          "len12_arg_accuracy": 0.0,
          "len12_program_exact": 0.0
        },
        {
          "stage": "answer_only",
          "local_step": 80,
          "step": 80,
          "loss": 3.7319068908691406,
          "trace_loss_active": false,
          "executor_loss": 3.7319068908691406,
          "len3_direct_accuracy": NaN,
          "len3_direct_target_mass": NaN,
          "len3_executor_accuracy": 0.0,
          "len3_executor_target_mass": 0.0089696254581213,
          "len3_init_accuracy": 0.0,
          "len3_op_accuracy": 0.3177083333333333,
          "len3_arg_accuracy": 0.0,
          "len3_program_exact": 0.0,
          "len6_direct_accuracy": NaN,
          "len6_direct_target_mass": NaN,
          "len6_executor_accuracy": 0.046875,
          "len6_executor_target_mass": 0.010430637747049332,
          "len6_init_accuracy": 0.0,
          "len6_op_accuracy": 0.2994791666666667,
          "len6_arg_accuracy": 0.010416666666666666,
          "len6_program_exact": 0.0,
          "len12_direct_accuracy": NaN,
          "len12_direct_target_mass": NaN,
          "len12_executor_accuracy": 0.0,
          "len12_executor_target_mass": 0.010294659994542599,
          "len12_init_accuracy": 0.0,
          "len12_op_accuracy": 0.35546875,
          "len12_arg_accuracy": 0.013020833333333334,
          "len12_program_exact": 0.0
        },
        {
          "stage": "answer_only",
          "local_step": 160,
          "step": 160,
          "loss": 2.6925594806671143,
          "trace_loss_active": false,
          "executor_loss": 2.6925594806671143,
          "len3_direct_accuracy": NaN,
          "len3_direct_target_mass": NaN,
          "len3_executor_accuracy": 0.015625,
          "len3_executor_target_mass": 0.005679427646100521,
          "len3_init_accuracy": 0.0,
          "len3_op_accuracy": 0.3177083333333333,
          "len3_arg_accuracy": 0.005208333333333333,
          "len3_program_exact": 0.0,
          "len6_direct_accuracy": NaN,
          "len6_direct_target_mass": NaN,
          "len6_executor_accuracy": 0.015625,
          "len6_executor_target_mass": 0.011833436787128448,
          "len6_init_accuracy": 0.0,
          "len6_op_accuracy": 0.2994791666666667,
          "len6_arg_accuracy": 0.0078125,
          "len6_program_exact": 0.0,
          "len12_direct_accuracy": NaN,
          "len12_direct_target_mass": NaN,
          "len12_executor_accuracy": 0.015625,
          "len12_executor_target_mass": 0.011265167966485023,
          "len12_init_accuracy": 0.0,
          "len12_op_accuracy": 0.35546875,
          "len12_arg_accuracy": 0.00390625,
          "len12_program_exact": 0.0
        },
        {
          "stage": "answer_only",
          "local_step": 240,
          "step": 240,
          "loss": 1.8998584747314453,
          "trace_loss_active": false,
          "executor_loss": 1.8998584747314453,
          "len3_direct_accuracy": NaN,
          "len3_direct_target_mass": NaN,
          "len3_executor_accuracy": 0.0,
          "len3_executor_target_mass": 0.004449417814612389,
          "len3_init_accuracy": 0.0,
          "len3_op_accuracy": 0.3177083333333333,
          "len3_arg_accuracy": 0.010416666666666666,
          "len3_program_exact": 0.0,
          "len6_direct_accuracy": NaN,
          "len6_direct_target_mass": NaN,
          "len6_executor_accuracy": 0.015625,
          "len6_executor_target_mass": 0.009476032108068466,
          "len6_init_accuracy": 0.0,
          "len6_op_accuracy": 0.2994791666666667,
          "len6_arg_accuracy": 0.0078125,
          "len6_program_exact": 0.0,
          "len12_direct_accuracy": NaN,
          "len12_direct_target_mass": NaN,
          "len12_executor_accuracy": 0.0,
          "len12_executor_target_mass": 0.013392070308327675,
          "len12_init_accuracy": 0.0,
          "len12_op_accuracy": 0.35546875,
          "len12_arg_accuracy": 0.0078125,
          "len12_program_exact": 0.0
        },
        {
          "stage": "answer_only",
          "local_step": 320,
          "step": 320,
          "loss": 0.9123830795288086,
          "trace_loss_active": false,
          "executor_loss": 0.9123830795288086,
          "len3_direct_accuracy": NaN,
          "len3_direct_target_mass": NaN,
          "len3_executor_accuracy": 0.0,
          "len3_executor_target_mass": 0.005006098188459873,
          "len3_init_accuracy": 0.0,
          "len3_op_accuracy": 0.3177083333333333,
          "len3_arg_accuracy": 0.010416666666666666,
          "len3_program_exact": 0.0,
          "len6_direct_accuracy": NaN,
          "len6_direct_target_mass": NaN,
          "len6_executor_accuracy": 0.03125,
          "len6_executor_target_mass": 0.00920964777469635,
          "len6_init_accuracy": 0.0,
          "len6_op_accuracy": 0.2994791666666667,
          "len6_arg_accuracy": 0.0078125,
          "len6_program_exact": 0.0,
          "len12_direct_accuracy": NaN,
          "len12_direct_target_mass": NaN,
          "len12_executor_accuracy": 0.015625,
          "len12_executor_target_mass": 0.01685580611228943,
          "len12_init_accuracy": 0.0,
          "len12_op_accuracy": 0.35546875,
          "len12_arg_accuracy": 0.010416666666666666,
          "len12_program_exact": 0.0
        }
      ],
      "final_metrics": {
        "len3": {
          "variant": "compiler_answer_only",
          "n": 64.0,
          "direct_accuracy": NaN,
          "direct_target_mass": NaN,
          "executor_accuracy": 0.0,
          "executor_target_mass": 0.005006098188459873,
          "init_accuracy": 0.0,
          "op_accuracy": 0.3177083333333333,
          "arg_accuracy": 0.010416666666666666,
          "program_exact": 0.0
        },
        "len6": {
          "variant": "compiler_answer_only",
          "n": 64.0,
          "direct_accuracy": NaN,
          "direct_target_mass": NaN,
          "executor_accuracy": 0.03125,
          "executor_target_mass": 0.00920964777469635,
          "init_accuracy": 0.0,
          "op_accuracy": 0.2994791666666667,
          "arg_accuracy": 0.0078125,
          "program_exact": 0.0
        },
        "len12": {
          "variant": "compiler_answer_only",
          "n": 64.0,
          "direct_accuracy": NaN,
          "direct_target_mass": NaN,
          "executor_accuracy": 0.015625,
          "executor_target_mass": 0.01685580611228943,
          "init_accuracy": 0.0,
          "op_accuracy": 0.35546875,
          "arg_accuracy": 0.010416666666666666,
          "program_exact": 0.0
        }
      }
    },
    "compiler_trace_then_answer": {
      "variant": "compiler_trace_then_answer",
      "train_seconds": 14.747769117355347,
      "checkpoints": [
        "large_artifacts/qwen_trace_bootstrap_retention/checkpoints/pilot_qwen35_retention/compiler_trace_then_answer/checkpoint_final.pt"
      ],
      "train_log": [
        {
          "stage": "trace_bootstrap",
          "local_step": 1,
          "step": 1,
          "loss": 14.77742862701416,
          "trace_loss_active": true,
          "executor_loss": 4.56760311126709,
          "init_loss": 4.584685802459717,
          "op_loss": 1.0734975337982178,
          "arg_loss": 4.551642417907715,
          "len3_direct_accuracy": NaN,
          "len3_direct_target_mass": NaN,
          "len3_executor_accuracy": 0.03125,
          "len3_executor_target_mass": 0.010420242324471474,
          "len3_init_accuracy": 0.015625,
          "len3_op_accuracy": 0.3125,
          "len3_arg_accuracy": 0.052083333333333336,
          "len3_program_exact": 0.0,
          "len6_direct_accuracy": NaN,
          "len6_direct_target_mass": NaN,
          "len6_executor_accuracy": 0.0,
          "len6_executor_target_mass": 0.010310012847185135,
          "len6_init_accuracy": 0.015625,
          "len6_op_accuracy": 0.3072916666666667,
          "len6_arg_accuracy": 0.08072916666666667,
          "len6_program_exact": 0.0,
          "len12_direct_accuracy": NaN,
          "len12_direct_target_mass": NaN,
          "len12_executor_accuracy": 0.0,
          "len12_executor_target_mass": 0.01030859723687172,
          "len12_init_accuracy": 0.046875,
          "len12_op_accuracy": 0.3606770833333333,
          "len12_arg_accuracy": 0.08854166666666667,
          "len12_program_exact": 0.0
        },
        {
          "stage": "trace_bootstrap",
          "local_step": 80,
          "step": 80,
          "loss": 0.6417288184165955,
          "trace_loss_active": true,
          "executor_loss": 0.3240417242050171,
          "init_loss": 0.2832038104534149,
          "op_loss": 0.00041302587487734854,
          "arg_loss": 0.03407026082277298,
          "len3_direct_accuracy": NaN,
          "len3_direct_target_mass": NaN,
          "len3_executor_accuracy": 0.59375,
          "len3_executor_target_mass": 0.452084481716156,
          "len3_init_accuracy": 0.625,
          "len3_op_accuracy": 1.0,
          "len3_arg_accuracy": 0.984375,
          "len3_program_exact": 0.59375,
          "len6_direct_accuracy": NaN,
          "len6_direct_target_mass": NaN,
          "len6_executor_accuracy": 0.703125,
          "len6_executor_target_mass": 0.4324464201927185,
          "len6_init_accuracy": 0.859375,
          "len6_op_accuracy": 0.9739583333333334,
          "len6_arg_accuracy": 0.9895833333333334,
          "len6_program_exact": 0.703125,
          "len12_direct_accuracy": NaN,
          "len12_direct_target_mass": NaN,
          "len12_executor_accuracy": 0.234375,
          "len12_executor_target_mass": 0.09750275313854218,
          "len12_init_accuracy": 0.796875,
          "len12_op_accuracy": 0.9114583333333334,
          "len12_arg_accuracy": 0.98828125,
          "len12_program_exact": 0.21875
        },
        {
          "stage": "trace_bootstrap",
          "local_step": 160,
          "step": 160,
          "loss": 0.30090469121932983,
          "trace_loss_active": true,
          "executor_loss": 0.09932515025138855,
          "init_loss": 0.07127602398395538,
          "op_loss": 0.12713994085788727,
          "arg_loss": 0.0031635884661227465,
          "len3_direct_accuracy": NaN,
          "len3_direct_target_mass": NaN,
          "len3_executor_accuracy": 0.796875,
          "len3_executor_target_mass": 0.7619559168815613,
          "len3_init_accuracy": 0.828125,
          "len3_op_accuracy": 1.0,
          "len3_arg_accuracy": 0.9895833333333334,
          "len3_program_exact": 0.796875,
          "len6_direct_accuracy": NaN,
          "len6_direct_target_mass": NaN,
          "len6_executor_accuracy": 0.65625,
          "len6_executor_target_mass": 0.6117251515388489,
          "len6_init_accuracy": 0.9375,
          "len6_op_accuracy": 0.953125,
          "len6_arg_accuracy": 0.9947916666666666,
          "len6_program_exact": 0.65625,
          "len12_direct_accuracy": NaN,
          "len12_direct_target_mass": NaN,
          "len12_executor_accuracy": 0.3125,
          "len12_executor_target_mass": 0.2081030160188675,
          "len12_init_accuracy": 0.890625,
          "len12_op_accuracy": 0.9036458333333334,
          "len12_arg_accuracy": 0.9908854166666666,
          "len12_program_exact": 0.3125
        },
        {
          "stage": "answer_retention",
          "local_step": 1,
          "step": 161,
          "loss": 2.4467132091522217,
          "trace_loss_active": false,
          "executor_loss": 2.4467132091522217,
          "len3_direct_accuracy": NaN,
          "len3_direct_target_mass": NaN,
          "len3_executor_accuracy": 0.796875,
          "len3_executor_target_mass": 0.7503727674484253,
          "len3_init_accuracy": 0.828125,
          "len3_op_accuracy": 1.0,
          "len3_arg_accuracy": 0.984375,
          "len3_program_exact": 0.796875,
          "len6_direct_accuracy": NaN,
          "len6_direct_target_mass": NaN,
          "len6_executor_accuracy": 0.59375,
          "len6_executor_target_mass": 0.5495160222053528,
          "len6_init_accuracy": 0.9375,
          "len6_op_accuracy": 0.9244791666666666,
          "len6_arg_accuracy": 0.9921875,
          "len6_program_exact": 0.59375,
          "len12_direct_accuracy": NaN,
          "len12_direct_target_mass": NaN,
          "len12_executor_accuracy": 0.109375,
          "len12_executor_target_mass": 0.10259201377630234,
          "len12_init_accuracy": 0.890625,
          "len12_op_accuracy": 0.8580729166666666,
          "len12_arg_accuracy": 0.9934895833333334,
          "len12_program_exact": 0.109375
        },
        {
          "stage": "answer_retention",
          "local_step": 80,
          "step": 240,
          "loss": 0.08221777528524399,
          "trace_loss_active": false,
          "executor_loss": 0.08221777528524399,
          "len3_direct_accuracy": NaN,
          "len3_direct_target_mass": NaN,
          "len3_executor_accuracy": 0.75,
          "len3_executor_target_mass": 0.7160650491714478,
          "len3_init_accuracy": 0.828125,
          "len3_op_accuracy": 1.0,
          "len3_arg_accuracy": 0.9739583333333334,
          "len3_program_exact": 0.75,
          "len6_direct_accuracy": NaN,
          "len6_direct_target_mass": NaN,
          "len6_executor_accuracy": 0.734375,
          "len6_executor_target_mass": 0.7048722505569458,
          "len6_init_accuracy": 0.9375,
          "len6_op_accuracy": 0.9817708333333334,
          "len6_arg_accuracy": 0.9817708333333334,
          "len6_program_exact": 0.734375,
          "len12_direct_accuracy": NaN,
          "len12_direct_target_mass": NaN,
          "len12_executor_accuracy": 0.546875,
          "len12_executor_target_mass": 0.44943296909332275,
          "len12_init_accuracy": 0.890625,
          "len12_op_accuracy": 0.984375,
          "len12_arg_accuracy": 0.9791666666666666,
          "len12_program_exact": 0.546875
        },
        {
          "stage": "answer_retention",
          "local_step": 160,
          "step": 320,
          "loss": 0.49726954102516174,
          "trace_loss_active": false,
          "executor_loss": 0.49726954102516174,
          "len3_direct_accuracy": NaN,
          "len3_direct_target_mass": NaN,
          "len3_executor_accuracy": 0.8125,
          "len3_executor_target_mass": 0.7555843591690063,
          "len3_init_accuracy": 0.828125,
          "len3_op_accuracy": 1.0,
          "len3_arg_accuracy": 0.9947916666666666,
          "len3_program_exact": 0.8125,
          "len6_direct_accuracy": NaN,
          "len6_direct_target_mass": NaN,
          "len6_executor_accuracy": 0.71875,
          "len6_executor_target_mass": 0.6945939064025879,
          "len6_init_accuracy": 0.9375,
          "len6_op_accuracy": 0.9869791666666666,
          "len6_arg_accuracy": 0.9713541666666666,
          "len6_program_exact": 0.71875,
          "len12_direct_accuracy": NaN,
          "len12_direct_target_mass": NaN,
          "len12_executor_accuracy": 0.5625,
          "len12_executor_target_mass": 0.4917149245738983,
          "len12_init_accuracy": 0.890625,
          "len12_op_accuracy": 0.9908854166666666,
          "len12_arg_accuracy": 0.9791666666666666,
          "len12_program_exact": 0.5625
        }
      ],
      "final_metrics": {
        "len3": {
          "variant": "compiler_trace_then_answer",
          "n": 64.0,
          "direct_accuracy": NaN,
          "direct_target_mass": NaN,
          "executor_accuracy": 0.8125,
          "executor_target_mass": 0.7555843591690063,
          "init_accuracy": 0.828125,
          "op_accuracy": 1.0,
          "arg_accuracy": 0.9947916666666666,
          "program_exact": 0.8125
        },
        "len6": {
          "variant": "compiler_trace_then_answer",
          "n": 64.0,
          "direct_accuracy": NaN,
          "direct_target_mass": NaN,
          "executor_accuracy": 0.71875,
          "executor_target_mass": 0.6945939064025879,
          "init_accuracy": 0.9375,
          "op_accuracy": 0.9869791666666666,
          "arg_accuracy": 0.9713541666666666,
          "program_exact": 0.71875
        },
        "len12": {
          "variant": "compiler_trace_then_answer",
          "n": 64.0,
          "direct_accuracy": NaN,
          "direct_target_mass": NaN,
          "executor_accuracy": 0.5625,
          "executor_target_mass": 0.4917149245738983,
          "init_accuracy": 0.890625,
          "op_accuracy": 0.9908854166666666,
          "arg_accuracy": 0.9791666666666666,
          "program_exact": 0.5625
        }
      }
    }
  }
}