{
  "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,
    "answer_lr_scale": 0.1,
    "weight_decay": 0.01,
    "grad_clip": 1.0,
    "head_width": 384,
    "trace_loss_weight": 1.0,
    "executor_loss_weight": 1.0,
    "variants": "compiler_trace_then_answer,compiler_trace_then_answer_low_lr",
    "max_length": 512,
    "seed": 7,
    "output_dir": "experiments/qwen_trace_bootstrap_retention/runs/pilot_qwen35_retention_lr",
    "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": {
    "compiler_trace_then_answer": {
      "variant": "compiler_trace_then_answer",
      "train_seconds": 14.960314989089966,
      "checkpoints": [
        "large_artifacts/qwen_trace_bootstrap_retention/checkpoints/pilot_qwen35_retention_lr/compiler_trace_then_answer/checkpoint_final.pt"
      ],
      "train_log": [
        {
          "stage": "trace_bootstrap",
          "local_step": 1,
          "step": 1,
          "loss": 14.860584259033203,
          "trace_loss_active": true,
          "lr_scale": 1.0,
          "executor_loss": 4.572367191314697,
          "init_loss": 4.610898017883301,
          "op_loss": 1.0667556524276733,
          "arg_loss": 4.610562801361084,
          "len3_direct_accuracy": NaN,
          "len3_direct_target_mass": NaN,
          "len3_executor_accuracy": 0.0,
          "len3_executor_target_mass": 0.010302882641553879,
          "len3_init_accuracy": 0.015625,
          "len3_op_accuracy": 0.3125,
          "len3_arg_accuracy": 0.0625,
          "len3_program_exact": 0.0,
          "len6_direct_accuracy": NaN,
          "len6_direct_target_mass": NaN,
          "len6_executor_accuracy": 0.015625,
          "len6_executor_target_mass": 0.010311109945178032,
          "len6_init_accuracy": 0.0,
          "len6_op_accuracy": 0.2994791666666667,
          "len6_arg_accuracy": 0.08333333333333333,
          "len6_program_exact": 0.0,
          "len12_direct_accuracy": NaN,
          "len12_direct_target_mass": NaN,
          "len12_executor_accuracy": 0.015625,
          "len12_executor_target_mass": 0.010307452641427517,
          "len12_init_accuracy": 0.0,
          "len12_op_accuracy": 0.3515625,
          "len12_arg_accuracy": 0.0625,
          "len12_program_exact": 0.0
        },
        {
          "stage": "trace_bootstrap",
          "local_step": 80,
          "step": 80,
          "loss": 0.4457215666770935,
          "trace_loss_active": true,
          "lr_scale": 1.0,
          "executor_loss": 0.2301473617553711,
          "init_loss": 0.19831618666648865,
          "op_loss": 0.0003176581230945885,
          "arg_loss": 0.01694035902619362,
          "len3_direct_accuracy": NaN,
          "len3_direct_target_mass": NaN,
          "len3_executor_accuracy": 0.6875,
          "len3_executor_target_mass": 0.4937804341316223,
          "len3_init_accuracy": 0.71875,
          "len3_op_accuracy": 1.0,
          "len3_arg_accuracy": 0.9895833333333334,
          "len3_program_exact": 0.6875,
          "len6_direct_accuracy": NaN,
          "len6_direct_target_mass": NaN,
          "len6_executor_accuracy": 0.734375,
          "len6_executor_target_mass": 0.46185651421546936,
          "len6_init_accuracy": 0.84375,
          "len6_op_accuracy": 0.984375,
          "len6_arg_accuracy": 0.9895833333333334,
          "len6_program_exact": 0.71875,
          "len12_direct_accuracy": NaN,
          "len12_direct_target_mass": NaN,
          "len12_executor_accuracy": 0.375,
          "len12_executor_target_mass": 0.17679807543754578,
          "len12_init_accuracy": 0.796875,
          "len12_op_accuracy": 0.94921875,
          "len12_arg_accuracy": 0.9934895833333334,
          "len12_program_exact": 0.375
        },
        {
          "stage": "trace_bootstrap",
          "local_step": 160,
          "step": 160,
          "loss": 0.056645192205905914,
          "trace_loss_active": true,
          "lr_scale": 1.0,
          "executor_loss": 0.03702555596828461,
          "init_loss": 0.002480903873220086,
          "op_loss": 0.016563652083277702,
          "arg_loss": 0.0005750794662162662,
          "len3_direct_accuracy": NaN,
          "len3_direct_target_mass": NaN,
          "len3_executor_accuracy": 0.8125,
          "len3_executor_target_mass": 0.7787091732025146,
          "len3_init_accuracy": 0.828125,
          "len3_op_accuracy": 0.9947916666666666,
          "len3_arg_accuracy": 0.9947916666666666,
          "len3_program_exact": 0.8125,
          "len6_direct_accuracy": NaN,
          "len6_direct_target_mass": NaN,
          "len6_executor_accuracy": 0.546875,
          "len6_executor_target_mass": 0.47947004437446594,
          "len6_init_accuracy": 0.9375,
          "len6_op_accuracy": 0.9140625,
          "len6_arg_accuracy": 1.0,
          "len6_program_exact": 0.546875,
          "len12_direct_accuracy": NaN,
          "len12_direct_target_mass": NaN,
          "len12_executor_accuracy": 0.03125,
          "len12_executor_target_mass": 0.02764570713043213,
          "len12_init_accuracy": 0.890625,
          "len12_op_accuracy": 0.7838541666666666,
          "len12_arg_accuracy": 0.9986979166666666,
          "len12_program_exact": 0.03125
        },
        {
          "stage": "answer_retention",
          "local_step": 1,
          "step": 161,
          "loss": 2.0820446014404297,
          "trace_loss_active": false,
          "lr_scale": 1.0,
          "executor_loss": 2.0820446014404297,
          "len3_direct_accuracy": NaN,
          "len3_direct_target_mass": NaN,
          "len3_executor_accuracy": 0.796875,
          "len3_executor_target_mass": 0.778535008430481,
          "len3_init_accuracy": 0.828125,
          "len3_op_accuracy": 0.9947916666666666,
          "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": 0.6004074811935425,
          "len6_init_accuracy": 0.9375,
          "len6_op_accuracy": 0.9375,
          "len6_arg_accuracy": 0.9973958333333334,
          "len6_program_exact": 0.65625,
          "len12_direct_accuracy": NaN,
          "len12_direct_target_mass": NaN,
          "len12_executor_accuracy": 0.171875,
          "len12_executor_target_mass": 0.13836584985256195,
          "len12_init_accuracy": 0.890625,
          "len12_op_accuracy": 0.8619791666666666,
          "len12_arg_accuracy": 0.99609375,
          "len12_program_exact": 0.171875
        },
        {
          "stage": "answer_retention",
          "local_step": 80,
          "step": 240,
          "loss": 0.17105762660503387,
          "trace_loss_active": false,
          "lr_scale": 1.0,
          "executor_loss": 0.17105762660503387,
          "len3_direct_accuracy": NaN,
          "len3_direct_target_mass": NaN,
          "len3_executor_accuracy": 0.796875,
          "len3_executor_target_mass": 0.7413195371627808,
          "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": 0.796875,
          "len6_executor_target_mass": 0.7069559693336487,
          "len6_init_accuracy": 0.9375,
          "len6_op_accuracy": 0.9973958333333334,
          "len6_arg_accuracy": 0.9791666666666666,
          "len6_program_exact": 0.796875,
          "len12_direct_accuracy": NaN,
          "len12_direct_target_mass": NaN,
          "len12_executor_accuracy": 0.640625,
          "len12_executor_target_mass": 0.4744895398616791,
          "len12_init_accuracy": 0.890625,
          "len12_op_accuracy": 0.9869791666666666,
          "len12_arg_accuracy": 0.9908854166666666,
          "len12_program_exact": 0.640625
        },
        {
          "stage": "answer_retention",
          "local_step": 160,
          "step": 320,
          "loss": 0.07883992791175842,
          "trace_loss_active": false,
          "lr_scale": 1.0,
          "executor_loss": 0.07883992791175842,
          "len3_direct_accuracy": NaN,
          "len3_direct_target_mass": NaN,
          "len3_executor_accuracy": 0.765625,
          "len3_executor_target_mass": 0.7387750148773193,
          "len3_init_accuracy": 0.828125,
          "len3_op_accuracy": 1.0,
          "len3_arg_accuracy": 0.9739583333333334,
          "len3_program_exact": 0.765625,
          "len6_direct_accuracy": NaN,
          "len6_direct_target_mass": NaN,
          "len6_executor_accuracy": 0.78125,
          "len6_executor_target_mass": 0.7257983684539795,
          "len6_init_accuracy": 0.9375,
          "len6_op_accuracy": 0.9973958333333334,
          "len6_arg_accuracy": 0.9713541666666666,
          "len6_program_exact": 0.78125,
          "len12_direct_accuracy": NaN,
          "len12_direct_target_mass": NaN,
          "len12_executor_accuracy": 0.5625,
          "len12_executor_target_mass": 0.5084843039512634,
          "len12_init_accuracy": 0.890625,
          "len12_op_accuracy": 0.9908854166666666,
          "len12_arg_accuracy": 0.9765625,
          "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": 0.765625,
          "executor_target_mass": 0.7387750148773193,
          "init_accuracy": 0.828125,
          "op_accuracy": 1.0,
          "arg_accuracy": 0.9739583333333334,
          "program_exact": 0.765625
        },
        "len6": {
          "variant": "compiler_trace_then_answer",
          "n": 64.0,
          "direct_accuracy": NaN,
          "direct_target_mass": NaN,
          "executor_accuracy": 0.78125,
          "executor_target_mass": 0.7257983684539795,
          "init_accuracy": 0.9375,
          "op_accuracy": 0.9973958333333334,
          "arg_accuracy": 0.9713541666666666,
          "program_exact": 0.78125
        },
        "len12": {
          "variant": "compiler_trace_then_answer",
          "n": 64.0,
          "direct_accuracy": NaN,
          "direct_target_mass": NaN,
          "executor_accuracy": 0.5625,
          "executor_target_mass": 0.5084843039512634,
          "init_accuracy": 0.890625,
          "op_accuracy": 0.9908854166666666,
          "arg_accuracy": 0.9765625,
          "program_exact": 0.5625
        }
      }
    },
    "compiler_trace_then_answer_low_lr": {
      "variant": "compiler_trace_then_answer_low_lr",
      "train_seconds": 14.287387132644653,
      "checkpoints": [
        "large_artifacts/qwen_trace_bootstrap_retention/checkpoints/pilot_qwen35_retention_lr/compiler_trace_then_answer_low_lr/checkpoint_final.pt"
      ],
      "train_log": [
        {
          "stage": "trace_bootstrap",
          "local_step": 1,
          "step": 1,
          "loss": 14.835021018981934,
          "trace_loss_active": true,
          "lr_scale": 1.0,
          "executor_loss": 4.559910774230957,
          "init_loss": 4.581078052520752,
          "op_loss": 1.0914976596832275,
          "arg_loss": 4.602534770965576,
          "len3_direct_accuracy": NaN,
          "len3_direct_target_mass": NaN,
          "len3_executor_accuracy": 0.0,
          "len3_executor_target_mass": 0.010297168977558613,
          "len3_init_accuracy": 0.0,
          "len3_op_accuracy": 0.34375,
          "len3_arg_accuracy": 0.0625,
          "len3_program_exact": 0.0,
          "len6_direct_accuracy": NaN,
          "len6_direct_target_mass": NaN,
          "len6_executor_accuracy": 0.03125,
          "len6_executor_target_mass": 0.010307282209396362,
          "len6_init_accuracy": 0.015625,
          "len6_op_accuracy": 0.3619791666666667,
          "len6_arg_accuracy": 0.049479166666666664,
          "len6_program_exact": 0.0,
          "len12_direct_accuracy": NaN,
          "len12_direct_target_mass": NaN,
          "len12_executor_accuracy": 0.0,
          "len12_executor_target_mass": 0.010306796059012413,
          "len12_init_accuracy": 0.0,
          "len12_op_accuracy": 0.3346354166666667,
          "len12_arg_accuracy": 0.0625,
          "len12_program_exact": 0.0
        },
        {
          "stage": "trace_bootstrap",
          "local_step": 80,
          "step": 80,
          "loss": 0.887605607509613,
          "trace_loss_active": true,
          "lr_scale": 1.0,
          "executor_loss": 0.4536522626876831,
          "init_loss": 0.4119054079055786,
          "op_loss": 0.00022817763965576887,
          "arg_loss": 0.02181977592408657,
          "len3_direct_accuracy": NaN,
          "len3_direct_target_mass": NaN,
          "len3_executor_accuracy": 0.71875,
          "len3_executor_target_mass": 0.5014615654945374,
          "len3_init_accuracy": 0.75,
          "len3_op_accuracy": 1.0,
          "len3_arg_accuracy": 0.9895833333333334,
          "len3_program_exact": 0.71875,
          "len6_direct_accuracy": NaN,
          "len6_direct_target_mass": NaN,
          "len6_executor_accuracy": 0.71875,
          "len6_executor_target_mass": 0.43794748187065125,
          "len6_init_accuracy": 0.859375,
          "len6_op_accuracy": 0.9817708333333334,
          "len6_arg_accuracy": 0.9921875,
          "len6_program_exact": 0.71875,
          "len12_direct_accuracy": NaN,
          "len12_direct_target_mass": NaN,
          "len12_executor_accuracy": 0.46875,
          "len12_executor_target_mass": 0.17439380288124084,
          "len12_init_accuracy": 0.765625,
          "len12_op_accuracy": 0.96484375,
          "len12_arg_accuracy": 0.9934895833333334,
          "len12_program_exact": 0.46875
        },
        {
          "stage": "trace_bootstrap",
          "local_step": 160,
          "step": 160,
          "loss": 0.0086351428180933,
          "trace_loss_active": true,
          "lr_scale": 1.0,
          "executor_loss": 0.00476333312690258,
          "init_loss": 0.0029046025592833757,
          "op_loss": 0.00018535478739067912,
          "arg_loss": 0.0007818524027243257,
          "len3_direct_accuracy": NaN,
          "len3_direct_target_mass": NaN,
          "len3_executor_accuracy": 0.78125,
          "len3_executor_target_mass": 0.7579014301300049,
          "len3_init_accuracy": 0.828125,
          "len3_op_accuracy": 0.9895833333333334,
          "len3_arg_accuracy": 0.9947916666666666,
          "len3_program_exact": 0.78125,
          "len6_direct_accuracy": NaN,
          "len6_direct_target_mass": NaN,
          "len6_executor_accuracy": 0.71875,
          "len6_executor_target_mass": 0.6827963590621948,
          "len6_init_accuracy": 0.9375,
          "len6_op_accuracy": 0.9609375,
          "len6_arg_accuracy": 0.9973958333333334,
          "len6_program_exact": 0.71875,
          "len12_direct_accuracy": NaN,
          "len12_direct_target_mass": NaN,
          "len12_executor_accuracy": 0.546875,
          "len12_executor_target_mass": 0.4454447329044342,
          "len12_init_accuracy": 0.890625,
          "len12_op_accuracy": 0.9596354166666666,
          "len12_arg_accuracy": 0.9986979166666666,
          "len12_program_exact": 0.546875
        },
        {
          "stage": "answer_retention_low_lr",
          "local_step": 1,
          "step": 161,
          "loss": 2.3472249507904053,
          "trace_loss_active": false,
          "lr_scale": 0.1,
          "executor_loss": 2.3472249507904053,
          "len3_direct_accuracy": NaN,
          "len3_direct_target_mass": NaN,
          "len3_executor_accuracy": 0.78125,
          "len3_executor_target_mass": 0.7598261833190918,
          "len3_init_accuracy": 0.828125,
          "len3_op_accuracy": 0.9895833333333334,
          "len3_arg_accuracy": 0.9947916666666666,
          "len3_program_exact": 0.78125,
          "len6_direct_accuracy": NaN,
          "len6_direct_target_mass": NaN,
          "len6_executor_accuracy": 0.71875,
          "len6_executor_target_mass": 0.6848634481430054,
          "len6_init_accuracy": 0.9375,
          "len6_op_accuracy": 0.9609375,
          "len6_arg_accuracy": 0.9973958333333334,
          "len6_program_exact": 0.71875,
          "len12_direct_accuracy": NaN,
          "len12_direct_target_mass": NaN,
          "len12_executor_accuracy": 0.546875,
          "len12_executor_target_mass": 0.44647955894470215,
          "len12_init_accuracy": 0.890625,
          "len12_op_accuracy": 0.9609375,
          "len12_arg_accuracy": 0.9986979166666666,
          "len12_program_exact": 0.546875
        },
        {
          "stage": "answer_retention_low_lr",
          "local_step": 80,
          "step": 240,
          "loss": 0.31240448355674744,
          "trace_loss_active": false,
          "lr_scale": 0.1,
          "executor_loss": 0.31240448355674744,
          "len3_direct_accuracy": NaN,
          "len3_direct_target_mass": NaN,
          "len3_executor_accuracy": 0.8125,
          "len3_executor_target_mass": 0.7520409226417542,
          "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": 0.75,
          "len6_executor_target_mass": 0.6132054924964905,
          "len6_init_accuracy": 0.9375,
          "len6_op_accuracy": 0.9739583333333334,
          "len6_arg_accuracy": 0.9869791666666666,
          "len6_program_exact": 0.75,
          "len12_direct_accuracy": NaN,
          "len12_direct_target_mass": NaN,
          "len12_executor_accuracy": 0.5,
          "len12_executor_target_mass": 0.2990996539592743,
          "len12_init_accuracy": 0.890625,
          "len12_op_accuracy": 0.9622395833333334,
          "len12_arg_accuracy": 0.9856770833333334,
          "len12_program_exact": 0.484375
        },
        {
          "stage": "answer_retention_low_lr",
          "local_step": 160,
          "step": 320,
          "loss": 0.2186330258846283,
          "trace_loss_active": false,
          "lr_scale": 0.1,
          "executor_loss": 0.2186330258846283,
          "len3_direct_accuracy": NaN,
          "len3_direct_target_mass": NaN,
          "len3_executor_accuracy": 0.796875,
          "len3_executor_target_mass": 0.7394669651985168,
          "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": 0.734375,
          "len6_executor_target_mass": 0.61106276512146,
          "len6_init_accuracy": 0.9375,
          "len6_op_accuracy": 0.984375,
          "len6_arg_accuracy": 0.9791666666666666,
          "len6_program_exact": 0.734375,
          "len12_direct_accuracy": NaN,
          "len12_direct_target_mass": NaN,
          "len12_executor_accuracy": 0.59375,
          "len12_executor_target_mass": 0.33492588996887207,
          "len12_init_accuracy": 0.890625,
          "len12_op_accuracy": 0.9739583333333334,
          "len12_arg_accuracy": 0.98828125,
          "len12_program_exact": 0.578125
        }
      ],
      "final_metrics": {
        "len3": {
          "variant": "compiler_trace_then_answer_low_lr",
          "n": 64.0,
          "direct_accuracy": NaN,
          "direct_target_mass": NaN,
          "executor_accuracy": 0.796875,
          "executor_target_mass": 0.7394669651985168,
          "init_accuracy": 0.828125,
          "op_accuracy": 1.0,
          "arg_accuracy": 0.984375,
          "program_exact": 0.796875
        },
        "len6": {
          "variant": "compiler_trace_then_answer_low_lr",
          "n": 64.0,
          "direct_accuracy": NaN,
          "direct_target_mass": NaN,
          "executor_accuracy": 0.734375,
          "executor_target_mass": 0.61106276512146,
          "init_accuracy": 0.9375,
          "op_accuracy": 0.984375,
          "arg_accuracy": 0.9791666666666666,
          "program_exact": 0.734375
        },
        "len12": {
          "variant": "compiler_trace_then_answer_low_lr",
          "n": 64.0,
          "direct_accuracy": NaN,
          "direct_target_mass": NaN,
          "executor_accuracy": 0.59375,
          "executor_target_mass": 0.33492588996887207,
          "init_accuracy": 0.890625,
          "op_accuracy": 0.9739583333333334,
          "arg_accuracy": 0.98828125,
          "program_exact": 0.578125
        }
      }
    }
  }
}