{
  "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": 160,
    "eval_size": 64,
    "feature_batch_size": 4,
    "train_batch_size": 64,
    "eval_batch_size": 64,
    "train_steps": 240,
    "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",
    "max_length": 512,
    "seed": 7,
    "output_dir": "experiments/qwen_structured_bridge/runs/pilot_qwen35_numeric_spans",
    "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": 160,
    "eval_size": 64,
    "train_lengths": [
      1,
      3
    ],
    "eval_lengths": [
      3,
      6,
      8
    ]
  },
  "variants": {
    "direct": {
      "variant": "direct",
      "train_seconds": 4.029103517532349,
      "checkpoints": [
        "large_artifacts/qwen_structured_bridge/checkpoints/pilot_qwen35_numeric_spans/direct/checkpoint_final.pt"
      ],
      "train_log": [
        {
          "step": 1,
          "loss": 4.6017279624938965,
          "direct_loss": 4.6017279624938965,
          "len3_direct_accuracy": 0.0,
          "len3_direct_target_mass": 0.007302397862076759,
          "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": 0.011128914542496204,
          "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": 0.009280472993850708,
          "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": 3.4034600257873535,
          "direct_loss": 3.4034600257873535,
          "len3_direct_accuracy": 0.015625,
          "len3_direct_target_mass": 0.010576671920716763,
          "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": 0.013568520545959473,
          "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": 0.03125,
          "len8_direct_target_mass": 0.01207856647670269,
          "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": 160,
          "loss": 2.9683244228363037,
          "direct_loss": 2.9683244228363037,
          "len3_direct_accuracy": 0.0,
          "len3_direct_target_mass": 0.010974356904625893,
          "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": 0.009055627509951591,
          "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": 0.009697295725345612,
          "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": 240,
          "loss": 2.4815330505371094,
          "direct_loss": 2.4815330505371094,
          "len3_direct_accuracy": 0.0,
          "len3_direct_target_mass": 0.009047867730259895,
          "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": 0.004435340408235788,
          "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": 0.007129944395273924,
          "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": 64.0,
          "direct_accuracy": 0.0,
          "direct_target_mass": 0.009047867730259895,
          "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": 0.004435340408235788,
          "executor_accuracy": NaN,
          "executor_target_mass": NaN,
          "init_accuracy": NaN,
          "op_accuracy": NaN,
          "arg_accuracy": NaN,
          "program_exact": NaN
        },
        "len8": {
          "variant": "direct",
          "n": 64.0,
          "direct_accuracy": 0.0,
          "direct_target_mass": 0.007129944395273924,
          "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": 9.66020917892456,
      "checkpoints": [
        "large_artifacts/qwen_structured_bridge/checkpoints/pilot_qwen35_numeric_spans/compiler_trace/checkpoint_final.pt"
      ],
      "train_log": [
        {
          "step": 1,
          "loss": 14.860601425170898,
          "executor_loss": 4.553466320037842,
          "init_loss": 4.597792148590088,
          "op_loss": 1.111067533493042,
          "arg_loss": 4.5982747077941895,
          "len3_direct_accuracy": NaN,
          "len3_direct_target_mass": NaN,
          "len3_executor_accuracy": 0.0,
          "len3_executor_target_mass": 0.01031329296529293,
          "len3_init_accuracy": 0.0,
          "len3_op_accuracy": 0.3177083333333333,
          "len3_arg_accuracy": 0.08854166666666667,
          "len3_program_exact": 0.0,
          "len6_direct_accuracy": NaN,
          "len6_direct_target_mass": NaN,
          "len6_executor_accuracy": 0.015625,
          "len6_executor_target_mass": 0.010308518074452877,
          "len6_init_accuracy": 0.015625,
          "len6_op_accuracy": 0.3020833333333333,
          "len6_arg_accuracy": 0.0546875,
          "len6_program_exact": 0.0,
          "len8_direct_accuracy": NaN,
          "len8_direct_target_mass": NaN,
          "len8_executor_accuracy": 0.03125,
          "len8_executor_target_mass": 0.010308986529707909,
          "len8_init_accuracy": 0.0,
          "len8_op_accuracy": 0.322265625,
          "len8_arg_accuracy": 0.033203125,
          "len8_program_exact": 0.0
        },
        {
          "step": 80,
          "loss": 0.48709797859191895,
          "executor_loss": 0.24934804439544678,
          "init_loss": 0.2194719910621643,
          "op_loss": 0.00010467042011441663,
          "arg_loss": 0.018173281103372574,
          "len3_direct_accuracy": NaN,
          "len3_direct_target_mass": NaN,
          "len3_executor_accuracy": 0.5625,
          "len3_executor_target_mass": 0.4434507489204407,
          "len3_init_accuracy": 0.609375,
          "len3_op_accuracy": 1.0,
          "len3_arg_accuracy": 0.9739583333333334,
          "len3_program_exact": 0.5625,
          "len6_direct_accuracy": NaN,
          "len6_direct_target_mass": NaN,
          "len6_executor_accuracy": 0.671875,
          "len6_executor_target_mass": 0.3930934965610504,
          "len6_init_accuracy": 0.828125,
          "len6_op_accuracy": 0.9869791666666666,
          "len6_arg_accuracy": 0.9739583333333334,
          "len6_program_exact": 0.65625,
          "len8_direct_accuracy": NaN,
          "len8_direct_target_mass": NaN,
          "len8_executor_accuracy": 0.53125,
          "len8_executor_target_mass": 0.29470008611679077,
          "len8_init_accuracy": 0.734375,
          "len8_op_accuracy": 0.978515625,
          "len8_arg_accuracy": 0.98046875,
          "len8_program_exact": 0.515625
        },
        {
          "step": 160,
          "loss": 0.006930043455213308,
          "executor_loss": 0.003727668896317482,
          "init_loss": 0.00271502323448658,
          "op_loss": 1.2462421182135586e-05,
          "arg_loss": 0.0004748889186885208,
          "len3_direct_accuracy": NaN,
          "len3_direct_target_mass": NaN,
          "len3_executor_accuracy": 0.71875,
          "len3_executor_target_mass": 0.6794718503952026,
          "len3_init_accuracy": 0.71875,
          "len3_op_accuracy": 1.0,
          "len3_arg_accuracy": 1.0,
          "len3_program_exact": 0.71875,
          "len6_direct_accuracy": NaN,
          "len6_direct_target_mass": NaN,
          "len6_executor_accuracy": 0.78125,
          "len6_executor_target_mass": 0.6851460337638855,
          "len6_init_accuracy": 0.859375,
          "len6_op_accuracy": 0.9921875,
          "len6_arg_accuracy": 0.9921875,
          "len6_program_exact": 0.78125,
          "len8_direct_accuracy": NaN,
          "len8_direct_target_mass": NaN,
          "len8_executor_accuracy": 0.703125,
          "len8_executor_target_mass": 0.5704529285430908,
          "len8_init_accuracy": 0.8125,
          "len8_op_accuracy": 0.98828125,
          "len8_arg_accuracy": 0.99609375,
          "len8_program_exact": 0.703125
        },
        {
          "step": 240,
          "loss": 0.0020679873414337635,
          "executor_loss": 0.0011066056322306395,
          "init_loss": 0.0007854477735236287,
          "op_loss": 3.913459295290522e-06,
          "arg_loss": 0.0001720203144941479,
          "len3_direct_accuracy": NaN,
          "len3_direct_target_mass": NaN,
          "len3_executor_accuracy": 0.71875,
          "len3_executor_target_mass": 0.6907600164413452,
          "len3_init_accuracy": 0.71875,
          "len3_op_accuracy": 1.0,
          "len3_arg_accuracy": 1.0,
          "len3_program_exact": 0.71875,
          "len6_direct_accuracy": NaN,
          "len6_direct_target_mass": NaN,
          "len6_executor_accuracy": 0.765625,
          "len6_executor_target_mass": 0.7204720973968506,
          "len6_init_accuracy": 0.859375,
          "len6_op_accuracy": 0.9895833333333334,
          "len6_arg_accuracy": 0.9947916666666666,
          "len6_program_exact": 0.765625,
          "len8_direct_accuracy": NaN,
          "len8_direct_target_mass": NaN,
          "len8_executor_accuracy": 0.71875,
          "len8_executor_target_mass": 0.6251443028450012,
          "len8_init_accuracy": 0.8125,
          "len8_op_accuracy": 0.98828125,
          "len8_arg_accuracy": 1.0,
          "len8_program_exact": 0.71875
        }
      ],
      "final_metrics": {
        "len3": {
          "variant": "compiler_trace",
          "n": 64.0,
          "direct_accuracy": NaN,
          "direct_target_mass": NaN,
          "executor_accuracy": 0.71875,
          "executor_target_mass": 0.6907600164413452,
          "init_accuracy": 0.71875,
          "op_accuracy": 1.0,
          "arg_accuracy": 1.0,
          "program_exact": 0.71875
        },
        "len6": {
          "variant": "compiler_trace",
          "n": 64.0,
          "direct_accuracy": NaN,
          "direct_target_mass": NaN,
          "executor_accuracy": 0.765625,
          "executor_target_mass": 0.7204720973968506,
          "init_accuracy": 0.859375,
          "op_accuracy": 0.9895833333333334,
          "arg_accuracy": 0.9947916666666666,
          "program_exact": 0.765625
        },
        "len8": {
          "variant": "compiler_trace",
          "n": 64.0,
          "direct_accuracy": NaN,
          "direct_target_mass": NaN,
          "executor_accuracy": 0.71875,
          "executor_target_mass": 0.6251443028450012,
          "init_accuracy": 0.8125,
          "op_accuracy": 0.98828125,
          "arg_accuracy": 1.0,
          "program_exact": 0.71875
        }
      }
    },
    "compiler_answer_only": {
      "variant": "compiler_answer_only",
      "train_seconds": 9.925841569900513,
      "checkpoints": [
        "large_artifacts/qwen_structured_bridge/checkpoints/pilot_qwen35_numeric_spans/compiler_answer_only/checkpoint_final.pt"
      ],
      "train_log": [
        {
          "step": 1,
          "loss": 4.569651126861572,
          "executor_loss": 4.569651126861572,
          "len3_direct_accuracy": NaN,
          "len3_direct_target_mass": NaN,
          "len3_executor_accuracy": 0.0,
          "len3_executor_target_mass": 2.5976873075705953e-05,
          "len3_init_accuracy": 0.03125,
          "len3_op_accuracy": 0.3958333333333333,
          "len3_arg_accuracy": 0.0,
          "len3_program_exact": 0.0,
          "len6_direct_accuracy": NaN,
          "len6_direct_target_mass": NaN,
          "len6_executor_accuracy": 0.015625,
          "len6_executor_target_mass": 0.01563114859163761,
          "len6_init_accuracy": 0.0,
          "len6_op_accuracy": 0.3489583333333333,
          "len6_arg_accuracy": 0.0,
          "len6_program_exact": 0.0,
          "len8_direct_accuracy": NaN,
          "len8_direct_target_mass": NaN,
          "len8_executor_accuracy": 0.015625,
          "len8_executor_target_mass": 0.015622738748788834,
          "len8_init_accuracy": 0.015625,
          "len8_op_accuracy": 0.341796875,
          "len8_arg_accuracy": 0.0,
          "len8_program_exact": 0.0
        },
        {
          "step": 80,
          "loss": 3.144416332244873,
          "executor_loss": 3.144416332244873,
          "len3_direct_accuracy": NaN,
          "len3_direct_target_mass": NaN,
          "len3_executor_accuracy": 0.0,
          "len3_executor_target_mass": 0.008606771938502789,
          "len3_init_accuracy": 0.03125,
          "len3_op_accuracy": 0.4010416666666667,
          "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": 0.009690498001873493,
          "len6_init_accuracy": 0.0,
          "len6_op_accuracy": 0.3567708333333333,
          "len6_arg_accuracy": 0.0026041666666666665,
          "len6_program_exact": 0.0,
          "len8_direct_accuracy": NaN,
          "len8_direct_target_mass": NaN,
          "len8_executor_accuracy": 0.0,
          "len8_executor_target_mass": 0.01065375842154026,
          "len8_init_accuracy": 0.015625,
          "len8_op_accuracy": 0.333984375,
          "len8_arg_accuracy": 0.0,
          "len8_program_exact": 0.0
        },
        {
          "step": 160,
          "loss": 1.478735327720642,
          "executor_loss": 1.478735327720642,
          "len3_direct_accuracy": NaN,
          "len3_direct_target_mass": NaN,
          "len3_executor_accuracy": 0.0,
          "len3_executor_target_mass": 0.009130634367465973,
          "len3_init_accuracy": 0.03125,
          "len3_op_accuracy": 0.4010416666666667,
          "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": 0.009622005745768547,
          "len6_init_accuracy": 0.0,
          "len6_op_accuracy": 0.3567708333333333,
          "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": 0.010210935026407242,
          "len8_init_accuracy": 0.015625,
          "len8_op_accuracy": 0.333984375,
          "len8_arg_accuracy": 0.00390625,
          "len8_program_exact": 0.0
        },
        {
          "step": 240,
          "loss": 0.9245887994766235,
          "executor_loss": 0.9245887994766235,
          "len3_direct_accuracy": NaN,
          "len3_direct_target_mass": NaN,
          "len3_executor_accuracy": 0.0,
          "len3_executor_target_mass": 0.010989965870976448,
          "len3_init_accuracy": 0.03125,
          "len3_op_accuracy": 0.4010416666666667,
          "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": 0.010756934061646461,
          "len6_init_accuracy": 0.0,
          "len6_op_accuracy": 0.3567708333333333,
          "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": 0.009538260288536549,
          "len8_init_accuracy": 0.015625,
          "len8_op_accuracy": 0.333984375,
          "len8_arg_accuracy": 0.001953125,
          "len8_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": 0.010989965870976448,
          "init_accuracy": 0.03125,
          "op_accuracy": 0.4010416666666667,
          "arg_accuracy": 0.0,
          "program_exact": 0.0
        },
        "len6": {
          "variant": "compiler_answer_only",
          "n": 64.0,
          "direct_accuracy": NaN,
          "direct_target_mass": NaN,
          "executor_accuracy": 0.046875,
          "executor_target_mass": 0.010756934061646461,
          "init_accuracy": 0.0,
          "op_accuracy": 0.3567708333333333,
          "arg_accuracy": 0.0,
          "program_exact": 0.0
        },
        "len8": {
          "variant": "compiler_answer_only",
          "n": 64.0,
          "direct_accuracy": NaN,
          "direct_target_mass": NaN,
          "executor_accuracy": 0.0,
          "executor_target_mass": 0.009538260288536549,
          "init_accuracy": 0.015625,
          "op_accuracy": 0.333984375,
          "arg_accuracy": 0.001953125,
          "program_exact": 0.0
        }
      }
    }
  }
}