{
  "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": 8,
    "train_min_len": 1,
    "train_max_len": 3,
    "eval_lengths": "3,6,8",
    "train_size": 96,
    "eval_size": 48,
    "feature_batch_size": 4,
    "train_batch_size": 32,
    "eval_batch_size": 64,
    "train_steps": 120,
    "eval_every": 40,
    "lr": 0.001,
    "weight_decay": 0.01,
    "grad_clip": 1.0,
    "head_width": 256,
    "trace_loss_weight": 1.0,
    "executor_loss_weight": 1.0,
    "variants": "direct,compiler_trace,compiler_answer_only",
    "max_length": 512,
    "seed": 7,
    "output_dir": "experiments/qwen_structured_bridge/runs/pilot_qwen35_frozen_bridge",
    "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": {
    "train_size": 96,
    "eval_size": 48,
    "train_lengths": [
      1,
      3
    ],
    "eval_lengths": [
      3,
      6,
      8
    ]
  },
  "variants": {
    "direct": {
      "variant": "direct",
      "train_seconds": 1.5022389888763428,
      "checkpoints": [
        "large_artifacts/qwen_structured_bridge/checkpoints/pilot_qwen35_frozen_bridge/direct/checkpoint_final.pt"
      ],
      "train_log": [
        {
          "step": 1,
          "loss": 4.63188362121582,
          "direct_loss": 4.63188362121582,
          "len3_direct_accuracy": 0.0,
          "len3_direct_target_mass": 0.014675735185543696,
          "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": 0.041666666666666664,
          "len6_direct_target_mass": 0.011885292828083038,
          "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,
          "len8_direct_accuracy": 0.0,
          "len8_direct_target_mass": 0.01178160309791565,
          "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
        },
        {
          "step": 40,
          "loss": 3.955873966217041,
          "direct_loss": 3.955873966217041,
          "len3_direct_accuracy": 0.020833333333333332,
          "len3_direct_target_mass": 0.012704862902561823,
          "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": 0.020833333333333332,
          "len6_direct_target_mass": 0.012154579162597656,
          "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,
          "len8_direct_accuracy": 0.0,
          "len8_direct_target_mass": 0.00982860599954923,
          "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
        },
        {
          "step": 80,
          "loss": 3.4222474098205566,
          "direct_loss": 3.4222474098205566,
          "len3_direct_accuracy": 0.020833333333333332,
          "len3_direct_target_mass": 0.010163570443789164,
          "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": 0.011646870523691177,
          "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,
          "len8_direct_accuracy": 0.020833333333333332,
          "len8_direct_target_mass": 0.013310181597868601,
          "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
        },
        {
          "step": 120,
          "loss": 2.889376401901245,
          "direct_loss": 2.889376401901245,
          "len3_direct_accuracy": 0.0,
          "len3_direct_target_mass": 0.009704011802872023,
          "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": 0.020833333333333332,
          "len6_direct_target_mass": 0.01205285390218099,
          "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,
          "len8_direct_accuracy": 0.0,
          "len8_direct_target_mass": 0.010832589119672775,
          "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
        }
      ],
      "final_metrics": {
        "len3": {
          "variant": "direct",
          "n": 48.0,
          "direct_accuracy": 0.0,
          "direct_target_mass": 0.009704011802872023,
          "executor_accuracy": NaN,
          "executor_target_mass": NaN,
          "init_accuracy": NaN,
          "op_accuracy": NaN,
          "arg_accuracy": NaN,
          "program_exact": NaN
        },
        "len6": {
          "variant": "direct",
          "n": 48.0,
          "direct_accuracy": 0.020833333333333332,
          "direct_target_mass": 0.01205285390218099,
          "executor_accuracy": NaN,
          "executor_target_mass": NaN,
          "init_accuracy": NaN,
          "op_accuracy": NaN,
          "arg_accuracy": NaN,
          "program_exact": NaN
        },
        "len8": {
          "variant": "direct",
          "n": 48.0,
          "direct_accuracy": 0.0,
          "direct_target_mass": 0.010832589119672775,
          "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": 4.05987548828125,
      "checkpoints": [
        "large_artifacts/qwen_structured_bridge/checkpoints/pilot_qwen35_frozen_bridge/compiler_trace/checkpoint_final.pt"
      ],
      "train_log": [
        {
          "step": 1,
          "loss": 14.855810165405273,
          "executor_loss": 4.557336807250977,
          "init_loss": 4.589353561401367,
          "op_loss": 1.1062953472137451,
          "arg_loss": 4.602823734283447,
          "len3_direct_accuracy": NaN,
          "len3_direct_target_mass": NaN,
          "len3_executor_accuracy": 0.041666666666666664,
          "len3_executor_target_mass": 0.010323333243529001,
          "len3_init_accuracy": 0.020833333333333332,
          "len3_op_accuracy": 0.3263888888888889,
          "len3_arg_accuracy": 0.09027777777777778,
          "len3_program_exact": 0.0,
          "len6_direct_accuracy": NaN,
          "len6_direct_target_mass": NaN,
          "len6_executor_accuracy": 0.0,
          "len6_executor_target_mass": 0.010309675087531408,
          "len6_init_accuracy": 0.020833333333333332,
          "len6_op_accuracy": 0.3229166666666667,
          "len6_arg_accuracy": 0.041666666666666664,
          "len6_program_exact": 0.0,
          "len8_direct_accuracy": NaN,
          "len8_direct_target_mass": NaN,
          "len8_executor_accuracy": 0.0,
          "len8_executor_target_mass": 0.010309378306070963,
          "len8_init_accuracy": 0.0,
          "len8_op_accuracy": 0.3333333333333333,
          "len8_arg_accuracy": 0.0546875,
          "len8_program_exact": 0.0
        },
        {
          "step": 40,
          "loss": 11.832118034362793,
          "executor_loss": 4.518031597137451,
          "init_loss": 4.005256652832031,
          "op_loss": 0.41164302825927734,
          "arg_loss": 2.897186756134033,
          "len3_direct_accuracy": NaN,
          "len3_direct_target_mass": NaN,
          "len3_executor_accuracy": 0.020833333333333332,
          "len3_executor_target_mass": 0.010265195742249489,
          "len3_init_accuracy": 0.0,
          "len3_op_accuracy": 0.8680555555555556,
          "len3_arg_accuracy": 0.06944444444444445,
          "len3_program_exact": 0.0,
          "len6_direct_accuracy": NaN,
          "len6_direct_target_mass": NaN,
          "len6_executor_accuracy": 0.041666666666666664,
          "len6_executor_target_mass": 0.010298818970719973,
          "len6_init_accuracy": 0.020833333333333332,
          "len6_op_accuracy": 0.7673611111111112,
          "len6_arg_accuracy": 0.04861111111111111,
          "len6_program_exact": 0.0,
          "len8_direct_accuracy": NaN,
          "len8_direct_target_mass": NaN,
          "len8_executor_accuracy": 0.0,
          "len8_executor_target_mass": 0.010297191018859545,
          "len8_init_accuracy": 0.020833333333333332,
          "len8_op_accuracy": 0.6979166666666666,
          "len8_arg_accuracy": 0.03125,
          "len8_program_exact": 0.0
        },
        {
          "step": 80,
          "loss": 10.737031936645508,
          "executor_loss": 4.4956817626953125,
          "init_loss": 3.9294304847717285,
          "op_loss": 0.08798206597566605,
          "arg_loss": 2.2239370346069336,
          "len3_direct_accuracy": NaN,
          "len3_direct_target_mass": NaN,
          "len3_executor_accuracy": 0.0,
          "len3_executor_target_mass": 0.010206634178757668,
          "len3_init_accuracy": 0.0,
          "len3_op_accuracy": 0.9583333333333334,
          "len3_arg_accuracy": 0.09722222222222222,
          "len3_program_exact": 0.0,
          "len6_direct_accuracy": NaN,
          "len6_direct_target_mass": NaN,
          "len6_executor_accuracy": 0.0,
          "len6_executor_target_mass": 0.010233216608564058,
          "len6_init_accuracy": 0.041666666666666664,
          "len6_op_accuracy": 0.875,
          "len6_arg_accuracy": 0.06944444444444445,
          "len6_program_exact": 0.0,
          "len8_direct_accuracy": NaN,
          "len8_direct_target_mass": NaN,
          "len8_executor_accuracy": 0.0,
          "len8_executor_target_mass": 0.010226496184865633,
          "len8_init_accuracy": 0.020833333333333332,
          "len8_op_accuracy": 0.90625,
          "len8_arg_accuracy": 0.06510416666666667,
          "len8_program_exact": 0.0
        },
        {
          "step": 120,
          "loss": 10.532458305358887,
          "executor_loss": 4.433683395385742,
          "init_loss": 4.0224409103393555,
          "op_loss": 0.03180306777358055,
          "arg_loss": 2.0445306301116943,
          "len3_direct_accuracy": NaN,
          "len3_direct_target_mass": NaN,
          "len3_executor_accuracy": 0.020833333333333332,
          "len3_executor_target_mass": 0.009987847258647283,
          "len3_init_accuracy": 0.0,
          "len3_op_accuracy": 0.9930555555555556,
          "len3_arg_accuracy": 0.14583333333333334,
          "len3_program_exact": 0.0,
          "len6_direct_accuracy": NaN,
          "len6_direct_target_mass": NaN,
          "len6_executor_accuracy": 0.0,
          "len6_executor_target_mass": 0.010021758576234182,
          "len6_init_accuracy": 0.041666666666666664,
          "len6_op_accuracy": 0.9965277777777778,
          "len6_arg_accuracy": 0.08680555555555555,
          "len6_program_exact": 0.0,
          "len8_direct_accuracy": NaN,
          "len8_direct_target_mass": NaN,
          "len8_executor_accuracy": 0.0,
          "len8_executor_target_mass": 0.010089298089345297,
          "len8_init_accuracy": 0.020833333333333332,
          "len8_op_accuracy": 0.9557291666666666,
          "len8_arg_accuracy": 0.09114583333333333,
          "len8_program_exact": 0.0
        }
      ],
      "final_metrics": {
        "len3": {
          "variant": "compiler_trace",
          "n": 48.0,
          "direct_accuracy": NaN,
          "direct_target_mass": NaN,
          "executor_accuracy": 0.020833333333333332,
          "executor_target_mass": 0.009987847258647283,
          "init_accuracy": 0.0,
          "op_accuracy": 0.9930555555555556,
          "arg_accuracy": 0.14583333333333334,
          "program_exact": 0.0
        },
        "len6": {
          "variant": "compiler_trace",
          "n": 48.0,
          "direct_accuracy": NaN,
          "direct_target_mass": NaN,
          "executor_accuracy": 0.0,
          "executor_target_mass": 0.010021758576234182,
          "init_accuracy": 0.041666666666666664,
          "op_accuracy": 0.9965277777777778,
          "arg_accuracy": 0.08680555555555555,
          "program_exact": 0.0
        },
        "len8": {
          "variant": "compiler_trace",
          "n": 48.0,
          "direct_accuracy": NaN,
          "direct_target_mass": NaN,
          "executor_accuracy": 0.0,
          "executor_target_mass": 0.010089298089345297,
          "init_accuracy": 0.020833333333333332,
          "op_accuracy": 0.9557291666666666,
          "arg_accuracy": 0.09114583333333333,
          "program_exact": 0.0
        }
      }
    },
    "compiler_answer_only": {
      "variant": "compiler_answer_only",
      "train_seconds": 3.9900519847869873,
      "checkpoints": [
        "large_artifacts/qwen_structured_bridge/checkpoints/pilot_qwen35_frozen_bridge/compiler_answer_only/checkpoint_final.pt"
      ],
      "train_log": [
        {
          "step": 1,
          "loss": 4.577580451965332,
          "executor_loss": 4.577580451965332,
          "len3_direct_accuracy": NaN,
          "len3_direct_target_mass": NaN,
          "len3_executor_accuracy": 0.0,
          "len3_executor_target_mass": 0.01031105654935042,
          "len3_init_accuracy": 0.041666666666666664,
          "len3_op_accuracy": 0.3263888888888889,
          "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": 0.010309278344114622,
          "len6_init_accuracy": 0.041666666666666664,
          "len6_op_accuracy": 0.3229166666666667,
          "len6_arg_accuracy": 0.0,
          "len6_program_exact": 0.0,
          "len8_direct_accuracy": NaN,
          "len8_direct_target_mass": NaN,
          "len8_executor_accuracy": 0.020833333333333332,
          "len8_executor_target_mass": 0.010309277723232904,
          "len8_init_accuracy": 0.0,
          "len8_op_accuracy": 0.3333333333333333,
          "len8_arg_accuracy": 0.0,
          "len8_program_exact": 0.0
        },
        {
          "step": 40,
          "loss": 4.343349456787109,
          "executor_loss": 4.343349456787109,
          "len3_direct_accuracy": NaN,
          "len3_direct_target_mass": NaN,
          "len3_executor_accuracy": 0.0,
          "len3_executor_target_mass": 0.010644649465878805,
          "len3_init_accuracy": 0.041666666666666664,
          "len3_op_accuracy": 0.3263888888888889,
          "len3_arg_accuracy": 0.020833333333333332,
          "len3_program_exact": 0.0,
          "len6_direct_accuracy": NaN,
          "len6_direct_target_mass": NaN,
          "len6_executor_accuracy": 0.0,
          "len6_executor_target_mass": 0.01030641794204712,
          "len6_init_accuracy": 0.0,
          "len6_op_accuracy": 0.3229166666666667,
          "len6_arg_accuracy": 0.013888888888888888,
          "len6_program_exact": 0.0,
          "len8_direct_accuracy": NaN,
          "len8_direct_target_mass": NaN,
          "len8_executor_accuracy": 0.020833333333333332,
          "len8_executor_target_mass": 0.010305254409710566,
          "len8_init_accuracy": 0.0,
          "len8_op_accuracy": 0.3333333333333333,
          "len8_arg_accuracy": 0.013020833333333334,
          "len8_program_exact": 0.0
        },
        {
          "step": 80,
          "loss": 4.510418891906738,
          "executor_loss": 4.510418891906738,
          "len3_direct_accuracy": NaN,
          "len3_direct_target_mass": NaN,
          "len3_executor_accuracy": 0.0625,
          "len3_executor_target_mass": 0.010850045830011368,
          "len3_init_accuracy": 0.0,
          "len3_op_accuracy": 0.3263888888888889,
          "len3_arg_accuracy": 0.013888888888888888,
          "len3_program_exact": 0.0,
          "len6_direct_accuracy": NaN,
          "len6_direct_target_mass": NaN,
          "len6_executor_accuracy": 0.0,
          "len6_executor_target_mass": 0.010598951329787573,
          "len6_init_accuracy": 0.041666666666666664,
          "len6_op_accuracy": 0.3229166666666667,
          "len6_arg_accuracy": 0.020833333333333332,
          "len6_program_exact": 0.0,
          "len8_direct_accuracy": NaN,
          "len8_direct_target_mass": NaN,
          "len8_executor_accuracy": 0.0,
          "len8_executor_target_mass": 0.010411584128936132,
          "len8_init_accuracy": 0.041666666666666664,
          "len8_op_accuracy": 0.3333333333333333,
          "len8_arg_accuracy": 0.020833333333333332,
          "len8_program_exact": 0.0
        },
        {
          "step": 120,
          "loss": 4.079866886138916,
          "executor_loss": 4.079866886138916,
          "len3_direct_accuracy": NaN,
          "len3_direct_target_mass": NaN,
          "len3_executor_accuracy": 0.020833333333333332,
          "len3_executor_target_mass": 0.01142919808626175,
          "len3_init_accuracy": 0.0,
          "len3_op_accuracy": 0.3263888888888889,
          "len3_arg_accuracy": 0.0,
          "len3_program_exact": 0.0,
          "len6_direct_accuracy": NaN,
          "len6_direct_target_mass": NaN,
          "len6_executor_accuracy": 0.020833333333333332,
          "len6_executor_target_mass": 0.010607520739237467,
          "len6_init_accuracy": 0.041666666666666664,
          "len6_op_accuracy": 0.3229166666666667,
          "len6_arg_accuracy": 0.0,
          "len6_program_exact": 0.0,
          "len8_direct_accuracy": NaN,
          "len8_direct_target_mass": NaN,
          "len8_executor_accuracy": 0.0,
          "len8_executor_target_mass": 0.010318034638961157,
          "len8_init_accuracy": 0.041666666666666664,
          "len8_op_accuracy": 0.3333333333333333,
          "len8_arg_accuracy": 0.005208333333333333,
          "len8_program_exact": 0.0
        }
      ],
      "final_metrics": {
        "len3": {
          "variant": "compiler_answer_only",
          "n": 48.0,
          "direct_accuracy": NaN,
          "direct_target_mass": NaN,
          "executor_accuracy": 0.020833333333333332,
          "executor_target_mass": 0.01142919808626175,
          "init_accuracy": 0.0,
          "op_accuracy": 0.3263888888888889,
          "arg_accuracy": 0.0,
          "program_exact": 0.0
        },
        "len6": {
          "variant": "compiler_answer_only",
          "n": 48.0,
          "direct_accuracy": NaN,
          "direct_target_mass": NaN,
          "executor_accuracy": 0.020833333333333332,
          "executor_target_mass": 0.010607520739237467,
          "init_accuracy": 0.041666666666666664,
          "op_accuracy": 0.3229166666666667,
          "arg_accuracy": 0.0,
          "program_exact": 0.0
        },
        "len8": {
          "variant": "compiler_answer_only",
          "n": 48.0,
          "direct_accuracy": NaN,
          "direct_target_mass": NaN,
          "executor_accuracy": 0.0,
          "executor_target_mass": 0.010318034638961157,
          "init_accuracy": 0.041666666666666664,
          "op_accuracy": 0.3333333333333333,
          "arg_accuracy": 0.005208333333333333,
          "program_exact": 0.0
        }
      }
    }
  }
}