{
  "args": {
    "model_id": "hf-internal-testing/tiny-random-LlamaForCausalLM",
    "text_model_fallback": "Qwen/Qwen3-4B",
    "model_class": "causal",
    "load_in_4bit": false,
    "torch_dtype": "fp32",
    "device_map": "auto",
    "modulus": 17,
    "max_steps": 6,
    "train_min_len": 1,
    "train_max_len": 2,
    "answer_train_min_len": 1,
    "answer_train_max_len": 4,
    "eval_lengths": "2,4",
    "train_template_mode": "standard",
    "answer_train_template_mode": "standard",
    "eval_template_modes": "standard,paraphrase",
    "train_size": 24,
    "answer_train_size": 24,
    "eval_size": 12,
    "feature_batch_size": 8,
    "train_batch_size": 8,
    "eval_batch_size": 16,
    "bootstrap_steps": 4,
    "answer_steps": 4,
    "eval_every": 4,
    "stage_eval_every": 2,
    "lr": 0.001,
    "answer_lr_scale": 0.1,
    "weight_decay": 0.01,
    "grad_clip": 1.0,
    "head_width": 64,
    "trace_loss_weight": 1.0,
    "executor_loss_weight": 1.0,
    "variants": "direct,compiler_trace,compiler_answer_only,compiler_trace_then_answer",
    "max_length": 512,
    "feature_seq_len": 128,
    "seed": 7,
    "output_dir": "experiments/qwen_span_free_compiler/runs/smoke_tiny",
    "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": "AutoModelForCausalLM",
    "peft_installed": true,
    "gpu_name": "NVIDIA RTX 6000 Ada Generation",
    "gpu_vram_gb": 47.383
  },
  "hidden_dim": 16,
  "dataset": {
    "bootstrap_train_size": 24,
    "answer_train_size": 24,
    "eval_size": 12,
    "bootstrap_train_lengths": [
      1,
      2
    ],
    "answer_train_lengths": [
      1,
      4
    ],
    "eval_lengths": [
      2,
      4
    ],
    "train_template_mode": "standard",
    "answer_train_template_mode": "standard",
    "eval_template_modes": [
      "standard",
      "paraphrase"
    ],
    "feature_seq_len": 128
  },
  "variants": {
    "direct": {
      "variant": "direct",
      "train_seconds": 0.22731423377990723,
      "checkpoints": [
        "large_artifacts/qwen_span_free_compiler/checkpoints/smoke_tiny/direct/checkpoint_final.pt"
      ],
      "train_log": [
        {
          "stage": "answer_only",
          "local_step": 1,
          "step": 1,
          "loss": 2.792112112045288,
          "trace_loss_active": false,
          "lr_scale": 1.0,
          "direct_loss": 2.792112112045288,
          "standard_len2_direct_accuracy": 0.08333333333333333,
          "standard_len2_direct_target_mass": 0.06300349036852519,
          "standard_len2_executor_accuracy": NaN,
          "standard_len2_executor_target_mass": NaN,
          "standard_len2_init_accuracy": NaN,
          "standard_len2_op_accuracy": NaN,
          "standard_len2_arg_accuracy": NaN,
          "standard_len2_program_exact": NaN,
          "standard_len4_direct_accuracy": 0.08333333333333333,
          "standard_len4_direct_target_mass": 0.05815687278906504,
          "standard_len4_executor_accuracy": NaN,
          "standard_len4_executor_target_mass": NaN,
          "standard_len4_init_accuracy": NaN,
          "standard_len4_op_accuracy": NaN,
          "standard_len4_arg_accuracy": NaN,
          "standard_len4_program_exact": NaN,
          "paraphrase_len2_direct_accuracy": 0.08333333333333333,
          "paraphrase_len2_direct_target_mass": 0.05778321623802185,
          "paraphrase_len2_executor_accuracy": NaN,
          "paraphrase_len2_executor_target_mass": NaN,
          "paraphrase_len2_init_accuracy": NaN,
          "paraphrase_len2_op_accuracy": NaN,
          "paraphrase_len2_arg_accuracy": NaN,
          "paraphrase_len2_program_exact": NaN,
          "paraphrase_len4_direct_accuracy": 0.08333333333333333,
          "paraphrase_len4_direct_target_mass": 0.05859164893627167,
          "paraphrase_len4_executor_accuracy": NaN,
          "paraphrase_len4_executor_target_mass": NaN,
          "paraphrase_len4_init_accuracy": NaN,
          "paraphrase_len4_op_accuracy": NaN,
          "paraphrase_len4_arg_accuracy": NaN,
          "paraphrase_len4_program_exact": NaN
        },
        {
          "stage": "answer_only",
          "local_step": 2,
          "step": 2,
          "loss": 2.857337474822998,
          "trace_loss_active": false,
          "lr_scale": 1.0,
          "direct_loss": 2.857337474822998,
          "standard_len2_direct_accuracy": 0.08333333333333333,
          "standard_len2_direct_target_mass": 0.06338492035865784,
          "standard_len2_executor_accuracy": NaN,
          "standard_len2_executor_target_mass": NaN,
          "standard_len2_init_accuracy": NaN,
          "standard_len2_op_accuracy": NaN,
          "standard_len2_arg_accuracy": NaN,
          "standard_len2_program_exact": NaN,
          "standard_len4_direct_accuracy": 0.08333333333333333,
          "standard_len4_direct_target_mass": 0.05797603726387024,
          "standard_len4_executor_accuracy": NaN,
          "standard_len4_executor_target_mass": NaN,
          "standard_len4_init_accuracy": NaN,
          "standard_len4_op_accuracy": NaN,
          "standard_len4_arg_accuracy": NaN,
          "standard_len4_program_exact": NaN,
          "paraphrase_len2_direct_accuracy": 0.08333333333333333,
          "paraphrase_len2_direct_target_mass": 0.05775047838687897,
          "paraphrase_len2_executor_accuracy": NaN,
          "paraphrase_len2_executor_target_mass": NaN,
          "paraphrase_len2_init_accuracy": NaN,
          "paraphrase_len2_op_accuracy": NaN,
          "paraphrase_len2_arg_accuracy": NaN,
          "paraphrase_len2_program_exact": NaN,
          "paraphrase_len4_direct_accuracy": 0.08333333333333333,
          "paraphrase_len4_direct_target_mass": 0.05893718699614207,
          "paraphrase_len4_executor_accuracy": NaN,
          "paraphrase_len4_executor_target_mass": NaN,
          "paraphrase_len4_init_accuracy": NaN,
          "paraphrase_len4_op_accuracy": NaN,
          "paraphrase_len4_arg_accuracy": NaN,
          "paraphrase_len4_program_exact": NaN
        },
        {
          "stage": "answer_only",
          "local_step": 4,
          "step": 4,
          "loss": 2.727606773376465,
          "trace_loss_active": false,
          "lr_scale": 1.0,
          "direct_loss": 2.727606773376465,
          "standard_len2_direct_accuracy": 0.08333333333333333,
          "standard_len2_direct_target_mass": 0.06387031078338623,
          "standard_len2_executor_accuracy": NaN,
          "standard_len2_executor_target_mass": NaN,
          "standard_len2_init_accuracy": NaN,
          "standard_len2_op_accuracy": NaN,
          "standard_len2_arg_accuracy": NaN,
          "standard_len2_program_exact": NaN,
          "standard_len4_direct_accuracy": 0.08333333333333333,
          "standard_len4_direct_target_mass": 0.05774283409118652,
          "standard_len4_executor_accuracy": NaN,
          "standard_len4_executor_target_mass": NaN,
          "standard_len4_init_accuracy": NaN,
          "standard_len4_op_accuracy": NaN,
          "standard_len4_arg_accuracy": NaN,
          "standard_len4_program_exact": NaN,
          "paraphrase_len2_direct_accuracy": 0.08333333333333333,
          "paraphrase_len2_direct_target_mass": 0.0575112650791804,
          "paraphrase_len2_executor_accuracy": NaN,
          "paraphrase_len2_executor_target_mass": NaN,
          "paraphrase_len2_init_accuracy": NaN,
          "paraphrase_len2_op_accuracy": NaN,
          "paraphrase_len2_arg_accuracy": NaN,
          "paraphrase_len2_program_exact": NaN,
          "paraphrase_len4_direct_accuracy": 0.08333333333333333,
          "paraphrase_len4_direct_target_mass": 0.05932555596033732,
          "paraphrase_len4_executor_accuracy": NaN,
          "paraphrase_len4_executor_target_mass": NaN,
          "paraphrase_len4_init_accuracy": NaN,
          "paraphrase_len4_op_accuracy": NaN,
          "paraphrase_len4_arg_accuracy": NaN,
          "paraphrase_len4_program_exact": NaN
        },
        {
          "stage": "answer_only",
          "local_step": 6,
          "step": 6,
          "loss": 2.7542505264282227,
          "trace_loss_active": false,
          "lr_scale": 1.0,
          "direct_loss": 2.7542505264282227,
          "standard_len2_direct_accuracy": 0.16666666666666666,
          "standard_len2_direct_target_mass": 0.06433513760566711,
          "standard_len2_executor_accuracy": NaN,
          "standard_len2_executor_target_mass": NaN,
          "standard_len2_init_accuracy": NaN,
          "standard_len2_op_accuracy": NaN,
          "standard_len2_arg_accuracy": NaN,
          "standard_len2_program_exact": NaN,
          "standard_len4_direct_accuracy": 0.0,
          "standard_len4_direct_target_mass": 0.057427167892456055,
          "standard_len4_executor_accuracy": NaN,
          "standard_len4_executor_target_mass": NaN,
          "standard_len4_init_accuracy": NaN,
          "standard_len4_op_accuracy": NaN,
          "standard_len4_arg_accuracy": NaN,
          "standard_len4_program_exact": NaN,
          "paraphrase_len2_direct_accuracy": 0.0,
          "paraphrase_len2_direct_target_mass": 0.057282855113347374,
          "paraphrase_len2_executor_accuracy": NaN,
          "paraphrase_len2_executor_target_mass": NaN,
          "paraphrase_len2_init_accuracy": NaN,
          "paraphrase_len2_op_accuracy": NaN,
          "paraphrase_len2_arg_accuracy": NaN,
          "paraphrase_len2_program_exact": NaN,
          "paraphrase_len4_direct_accuracy": 0.0,
          "paraphrase_len4_direct_target_mass": 0.059443081418673195,
          "paraphrase_len4_executor_accuracy": NaN,
          "paraphrase_len4_executor_target_mass": NaN,
          "paraphrase_len4_init_accuracy": NaN,
          "paraphrase_len4_op_accuracy": NaN,
          "paraphrase_len4_arg_accuracy": NaN,
          "paraphrase_len4_program_exact": NaN
        },
        {
          "stage": "answer_only",
          "local_step": 8,
          "step": 8,
          "loss": 2.696685791015625,
          "trace_loss_active": false,
          "lr_scale": 1.0,
          "direct_loss": 2.696685791015625,
          "standard_len2_direct_accuracy": 0.16666666666666666,
          "standard_len2_direct_target_mass": 0.06465814510981242,
          "standard_len2_executor_accuracy": NaN,
          "standard_len2_executor_target_mass": NaN,
          "standard_len2_init_accuracy": NaN,
          "standard_len2_op_accuracy": NaN,
          "standard_len2_arg_accuracy": NaN,
          "standard_len2_program_exact": NaN,
          "standard_len4_direct_accuracy": 0.0,
          "standard_len4_direct_target_mass": 0.05699672798315684,
          "standard_len4_executor_accuracy": NaN,
          "standard_len4_executor_target_mass": NaN,
          "standard_len4_init_accuracy": NaN,
          "standard_len4_op_accuracy": NaN,
          "standard_len4_arg_accuracy": NaN,
          "standard_len4_program_exact": NaN,
          "paraphrase_len2_direct_accuracy": 0.0,
          "paraphrase_len2_direct_target_mass": 0.0572198232014974,
          "paraphrase_len2_executor_accuracy": NaN,
          "paraphrase_len2_executor_target_mass": NaN,
          "paraphrase_len2_init_accuracy": NaN,
          "paraphrase_len2_op_accuracy": NaN,
          "paraphrase_len2_arg_accuracy": NaN,
          "paraphrase_len2_program_exact": NaN,
          "paraphrase_len4_direct_accuracy": 0.0,
          "paraphrase_len4_direct_target_mass": 0.059570337335268654,
          "paraphrase_len4_executor_accuracy": NaN,
          "paraphrase_len4_executor_target_mass": NaN,
          "paraphrase_len4_init_accuracy": NaN,
          "paraphrase_len4_op_accuracy": NaN,
          "paraphrase_len4_arg_accuracy": NaN,
          "paraphrase_len4_program_exact": NaN
        }
      ],
      "final_metrics": {
        "standard_len2": {
          "variant": "direct",
          "n": 12.0,
          "direct_accuracy": 0.16666666666666666,
          "direct_target_mass": 0.06465814510981242,
          "executor_accuracy": NaN,
          "executor_target_mass": NaN,
          "init_accuracy": NaN,
          "op_accuracy": NaN,
          "arg_accuracy": NaN,
          "program_exact": NaN
        },
        "standard_len4": {
          "variant": "direct",
          "n": 12.0,
          "direct_accuracy": 0.0,
          "direct_target_mass": 0.05699672798315684,
          "executor_accuracy": NaN,
          "executor_target_mass": NaN,
          "init_accuracy": NaN,
          "op_accuracy": NaN,
          "arg_accuracy": NaN,
          "program_exact": NaN
        },
        "paraphrase_len2": {
          "variant": "direct",
          "n": 12.0,
          "direct_accuracy": 0.0,
          "direct_target_mass": 0.0572198232014974,
          "executor_accuracy": NaN,
          "executor_target_mass": NaN,
          "init_accuracy": NaN,
          "op_accuracy": NaN,
          "arg_accuracy": NaN,
          "program_exact": NaN
        },
        "paraphrase_len4": {
          "variant": "direct",
          "n": 12.0,
          "direct_accuracy": 0.0,
          "direct_target_mass": 0.059570337335268654,
          "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": 0.9029409885406494,
      "checkpoints": [
        "large_artifacts/qwen_span_free_compiler/checkpoints/smoke_tiny/compiler_trace/checkpoint_final.pt"
      ],
      "train_log": [
        {
          "stage": "trace_bootstrap",
          "local_step": 1,
          "step": 1,
          "loss": 9.64088249206543,
          "trace_loss_active": true,
          "lr_scale": 1.0,
          "executor_loss": 2.8103556632995605,
          "init_loss": 2.8510940074920654,
          "op_loss": 1.0996371507644653,
          "arg_loss": 2.879795551300049,
          "standard_len2_direct_accuracy": NaN,
          "standard_len2_direct_target_mass": NaN,
          "standard_len2_executor_accuracy": 0.16666666666666666,
          "standard_len2_executor_target_mass": 0.062424282232920326,
          "standard_len2_init_accuracy": 0.3333333333333333,
          "standard_len2_op_accuracy": 0.4583333333333333,
          "standard_len2_arg_accuracy": 0.041666666666666664,
          "standard_len2_program_exact": 0.0,
          "standard_len4_direct_accuracy": NaN,
          "standard_len4_direct_target_mass": NaN,
          "standard_len4_executor_accuracy": 0.0,
          "standard_len4_executor_target_mass": 0.05719950795173645,
          "standard_len4_init_accuracy": 0.0,
          "standard_len4_op_accuracy": 0.375,
          "standard_len4_arg_accuracy": 0.0625,
          "standard_len4_program_exact": 0.0,
          "paraphrase_len2_direct_accuracy": NaN,
          "paraphrase_len2_direct_target_mass": NaN,
          "paraphrase_len2_executor_accuracy": 0.08333333333333333,
          "paraphrase_len2_executor_target_mass": 0.05985637505849203,
          "paraphrase_len2_init_accuracy": 0.25,
          "paraphrase_len2_op_accuracy": 0.3333333333333333,
          "paraphrase_len2_arg_accuracy": 0.041666666666666664,
          "paraphrase_len2_program_exact": 0.0,
          "paraphrase_len4_direct_accuracy": NaN,
          "paraphrase_len4_direct_target_mass": NaN,
          "paraphrase_len4_executor_accuracy": 0.0,
          "paraphrase_len4_executor_target_mass": 0.059288159012794495,
          "paraphrase_len4_init_accuracy": 0.0,
          "paraphrase_len4_op_accuracy": 0.2916666666666667,
          "paraphrase_len4_arg_accuracy": 0.041666666666666664,
          "paraphrase_len4_program_exact": 0.0
        },
        {
          "stage": "trace_bootstrap",
          "local_step": 2,
          "step": 2,
          "loss": 9.653790473937988,
          "trace_loss_active": true,
          "lr_scale": 1.0,
          "executor_loss": 2.867525339126587,
          "init_loss": 2.8197054862976074,
          "op_loss": 1.0619175434112549,
          "arg_loss": 2.904642105102539,
          "standard_len2_direct_accuracy": NaN,
          "standard_len2_direct_target_mass": NaN,
          "standard_len2_executor_accuracy": 0.16666666666666666,
          "standard_len2_executor_target_mass": 0.06245217720667521,
          "standard_len2_init_accuracy": 0.3333333333333333,
          "standard_len2_op_accuracy": 0.4583333333333333,
          "standard_len2_arg_accuracy": 0.041666666666666664,
          "standard_len2_program_exact": 0.0,
          "standard_len4_direct_accuracy": NaN,
          "standard_len4_direct_target_mass": NaN,
          "standard_len4_executor_accuracy": 0.0,
          "standard_len4_executor_target_mass": 0.057158450285593666,
          "standard_len4_init_accuracy": 0.0,
          "standard_len4_op_accuracy": 0.375,
          "standard_len4_arg_accuracy": 0.08333333333333333,
          "standard_len4_program_exact": 0.0,
          "paraphrase_len2_direct_accuracy": NaN,
          "paraphrase_len2_direct_target_mass": NaN,
          "paraphrase_len2_executor_accuracy": 0.08333333333333333,
          "paraphrase_len2_executor_target_mass": 0.05986723303794861,
          "paraphrase_len2_init_accuracy": 0.25,
          "paraphrase_len2_op_accuracy": 0.3333333333333333,
          "paraphrase_len2_arg_accuracy": 0.041666666666666664,
          "paraphrase_len2_program_exact": 0.0,
          "paraphrase_len4_direct_accuracy": NaN,
          "paraphrase_len4_direct_target_mass": NaN,
          "paraphrase_len4_executor_accuracy": 0.0,
          "paraphrase_len4_executor_target_mass": 0.05932582418123881,
          "paraphrase_len4_init_accuracy": 0.0,
          "paraphrase_len4_op_accuracy": 0.2916666666666667,
          "paraphrase_len4_arg_accuracy": 0.041666666666666664,
          "paraphrase_len4_program_exact": 0.0
        },
        {
          "stage": "trace_bootstrap",
          "local_step": 4,
          "step": 4,
          "loss": 9.623966217041016,
          "trace_loss_active": true,
          "lr_scale": 1.0,
          "executor_loss": 2.8640224933624268,
          "init_loss": 2.7527637481689453,
          "op_loss": 1.171026349067688,
          "arg_loss": 2.836153745651245,
          "standard_len2_direct_accuracy": NaN,
          "standard_len2_direct_target_mass": NaN,
          "standard_len2_executor_accuracy": 0.08333333333333333,
          "standard_len2_executor_target_mass": 0.061946362257003784,
          "standard_len2_init_accuracy": 0.0,
          "standard_len2_op_accuracy": 0.4583333333333333,
          "standard_len2_arg_accuracy": 0.041666666666666664,
          "standard_len2_program_exact": 0.0,
          "standard_len4_direct_accuracy": NaN,
          "standard_len4_direct_target_mass": NaN,
          "standard_len4_executor_accuracy": 0.0,
          "standard_len4_executor_target_mass": 0.05734856923421224,
          "standard_len4_init_accuracy": 0.0,
          "standard_len4_op_accuracy": 0.4166666666666667,
          "standard_len4_arg_accuracy": 0.0625,
          "standard_len4_program_exact": 0.0,
          "paraphrase_len2_direct_accuracy": NaN,
          "paraphrase_len2_direct_target_mass": NaN,
          "paraphrase_len2_executor_accuracy": 0.08333333333333333,
          "paraphrase_len2_executor_target_mass": 0.05975225567817688,
          "paraphrase_len2_init_accuracy": 0.25,
          "paraphrase_len2_op_accuracy": 0.3333333333333333,
          "paraphrase_len2_arg_accuracy": 0.041666666666666664,
          "paraphrase_len2_program_exact": 0.0,
          "paraphrase_len4_direct_accuracy": NaN,
          "paraphrase_len4_direct_target_mass": NaN,
          "paraphrase_len4_executor_accuracy": 0.08333333333333333,
          "paraphrase_len4_executor_target_mass": 0.05926594138145447,
          "paraphrase_len4_init_accuracy": 0.0,
          "paraphrase_len4_op_accuracy": 0.2708333333333333,
          "paraphrase_len4_arg_accuracy": 0.041666666666666664,
          "paraphrase_len4_program_exact": 0.0
        },
        {
          "stage": "trace_continuation",
          "local_step": 1,
          "step": 5,
          "loss": 9.671706199645996,
          "trace_loss_active": true,
          "lr_scale": 1.0,
          "executor_loss": 2.8143668174743652,
          "init_loss": 2.917365312576294,
          "op_loss": 1.0743904113769531,
          "arg_loss": 2.865583658218384,
          "standard_len2_direct_accuracy": NaN,
          "standard_len2_direct_target_mass": NaN,
          "standard_len2_executor_accuracy": 0.08333333333333333,
          "standard_len2_executor_target_mass": 0.06173346439997355,
          "standard_len2_init_accuracy": 0.0,
          "standard_len2_op_accuracy": 0.4583333333333333,
          "standard_len2_arg_accuracy": 0.041666666666666664,
          "standard_len2_program_exact": 0.0,
          "standard_len4_direct_accuracy": NaN,
          "standard_len4_direct_target_mass": NaN,
          "standard_len4_executor_accuracy": 0.08333333333333333,
          "standard_len4_executor_target_mass": 0.05741016070048014,
          "standard_len4_init_accuracy": 0.0,
          "standard_len4_op_accuracy": 0.4375,
          "standard_len4_arg_accuracy": 0.0625,
          "standard_len4_program_exact": 0.0,
          "paraphrase_len2_direct_accuracy": NaN,
          "paraphrase_len2_direct_target_mass": NaN,
          "paraphrase_len2_executor_accuracy": 0.08333333333333333,
          "paraphrase_len2_executor_target_mass": 0.05970896780490875,
          "paraphrase_len2_init_accuracy": 0.16666666666666666,
          "paraphrase_len2_op_accuracy": 0.375,
          "paraphrase_len2_arg_accuracy": 0.041666666666666664,
          "paraphrase_len2_program_exact": 0.0,
          "paraphrase_len4_direct_accuracy": NaN,
          "paraphrase_len4_direct_target_mass": NaN,
          "paraphrase_len4_executor_accuracy": 0.08333333333333333,
          "paraphrase_len4_executor_target_mass": 0.059247891108194985,
          "paraphrase_len4_init_accuracy": 0.0,
          "paraphrase_len4_op_accuracy": 0.2708333333333333,
          "paraphrase_len4_arg_accuracy": 0.0625,
          "paraphrase_len4_program_exact": 0.0
        },
        {
          "stage": "trace_continuation",
          "local_step": 2,
          "step": 6,
          "loss": 9.607881546020508,
          "trace_loss_active": true,
          "lr_scale": 1.0,
          "executor_loss": 2.8106613159179688,
          "init_loss": 2.9348154067993164,
          "op_loss": 1.0428025722503662,
          "arg_loss": 2.8196027278900146,
          "standard_len2_direct_accuracy": NaN,
          "standard_len2_direct_target_mass": NaN,
          "standard_len2_executor_accuracy": 0.08333333333333333,
          "standard_len2_executor_target_mass": 0.06169002254803976,
          "standard_len2_init_accuracy": 0.0,
          "standard_len2_op_accuracy": 0.4583333333333333,
          "standard_len2_arg_accuracy": 0.041666666666666664,
          "standard_len2_program_exact": 0.0,
          "standard_len4_direct_accuracy": NaN,
          "standard_len4_direct_target_mass": NaN,
          "standard_len4_executor_accuracy": 0.08333333333333333,
          "standard_len4_executor_target_mass": 0.05737613638242086,
          "standard_len4_init_accuracy": 0.0,
          "standard_len4_op_accuracy": 0.3541666666666667,
          "standard_len4_arg_accuracy": 0.08333333333333333,
          "standard_len4_program_exact": 0.0,
          "paraphrase_len2_direct_accuracy": NaN,
          "paraphrase_len2_direct_target_mass": NaN,
          "paraphrase_len2_executor_accuracy": 0.08333333333333333,
          "paraphrase_len2_executor_target_mass": 0.05971899628639221,
          "paraphrase_len2_init_accuracy": 0.0,
          "paraphrase_len2_op_accuracy": 0.3333333333333333,
          "paraphrase_len2_arg_accuracy": 0.041666666666666664,
          "paraphrase_len2_program_exact": 0.0,
          "paraphrase_len4_direct_accuracy": NaN,
          "paraphrase_len4_direct_target_mass": NaN,
          "paraphrase_len4_executor_accuracy": 0.0,
          "paraphrase_len4_executor_target_mass": 0.05925891796747843,
          "paraphrase_len4_init_accuracy": 0.0,
          "paraphrase_len4_op_accuracy": 0.2708333333333333,
          "paraphrase_len4_arg_accuracy": 0.0625,
          "paraphrase_len4_program_exact": 0.0
        },
        {
          "stage": "trace_continuation",
          "local_step": 4,
          "step": 8,
          "loss": 9.398048400878906,
          "trace_loss_active": true,
          "lr_scale": 1.0,
          "executor_loss": 2.856745719909668,
          "init_loss": 2.666212797164917,
          "op_loss": 1.0935249328613281,
          "arg_loss": 2.7815651893615723,
          "standard_len2_direct_accuracy": NaN,
          "standard_len2_direct_target_mass": NaN,
          "standard_len2_executor_accuracy": 0.0,
          "standard_len2_executor_target_mass": 0.06161861618359884,
          "standard_len2_init_accuracy": 0.0,
          "standard_len2_op_accuracy": 0.4583333333333333,
          "standard_len2_arg_accuracy": 0.041666666666666664,
          "standard_len2_program_exact": 0.0,
          "standard_len4_direct_accuracy": NaN,
          "standard_len4_direct_target_mass": NaN,
          "standard_len4_executor_accuracy": 0.0,
          "standard_len4_executor_target_mass": 0.05729359885056814,
          "standard_len4_init_accuracy": 0.0,
          "standard_len4_op_accuracy": 0.2708333333333333,
          "standard_len4_arg_accuracy": 0.08333333333333333,
          "standard_len4_program_exact": 0.0,
          "paraphrase_len2_direct_accuracy": NaN,
          "paraphrase_len2_direct_target_mass": NaN,
          "paraphrase_len2_executor_accuracy": 0.08333333333333333,
          "paraphrase_len2_executor_target_mass": 0.05971568822860718,
          "paraphrase_len2_init_accuracy": 0.0,
          "paraphrase_len2_op_accuracy": 0.3333333333333333,
          "paraphrase_len2_arg_accuracy": 0.041666666666666664,
          "paraphrase_len2_program_exact": 0.0,
          "paraphrase_len4_direct_accuracy": NaN,
          "paraphrase_len4_direct_target_mass": NaN,
          "paraphrase_len4_executor_accuracy": 0.25,
          "paraphrase_len4_executor_target_mass": 0.0593005915482839,
          "paraphrase_len4_init_accuracy": 0.08333333333333333,
          "paraphrase_len4_op_accuracy": 0.3541666666666667,
          "paraphrase_len4_arg_accuracy": 0.041666666666666664,
          "paraphrase_len4_program_exact": 0.0
        }
      ],
      "final_metrics": {
        "standard_len2": {
          "variant": "compiler_trace",
          "n": 12.0,
          "direct_accuracy": NaN,
          "direct_target_mass": NaN,
          "executor_accuracy": 0.0,
          "executor_target_mass": 0.06161861618359884,
          "init_accuracy": 0.0,
          "op_accuracy": 0.4583333333333333,
          "arg_accuracy": 0.041666666666666664,
          "program_exact": 0.0
        },
        "standard_len4": {
          "variant": "compiler_trace",
          "n": 12.0,
          "direct_accuracy": NaN,
          "direct_target_mass": NaN,
          "executor_accuracy": 0.0,
          "executor_target_mass": 0.05729359885056814,
          "init_accuracy": 0.0,
          "op_accuracy": 0.2708333333333333,
          "arg_accuracy": 0.08333333333333333,
          "program_exact": 0.0
        },
        "paraphrase_len2": {
          "variant": "compiler_trace",
          "n": 12.0,
          "direct_accuracy": NaN,
          "direct_target_mass": NaN,
          "executor_accuracy": 0.08333333333333333,
          "executor_target_mass": 0.05971568822860718,
          "init_accuracy": 0.0,
          "op_accuracy": 0.3333333333333333,
          "arg_accuracy": 0.041666666666666664,
          "program_exact": 0.0
        },
        "paraphrase_len4": {
          "variant": "compiler_trace",
          "n": 12.0,
          "direct_accuracy": NaN,
          "direct_target_mass": NaN,
          "executor_accuracy": 0.25,
          "executor_target_mass": 0.0593005915482839,
          "init_accuracy": 0.08333333333333333,
          "op_accuracy": 0.3541666666666667,
          "arg_accuracy": 0.041666666666666664,
          "program_exact": 0.0
        }
      }
    },
    "compiler_answer_only": {
      "variant": "compiler_answer_only",
      "train_seconds": 0.7715888023376465,
      "checkpoints": [
        "large_artifacts/qwen_span_free_compiler/checkpoints/smoke_tiny/compiler_answer_only/checkpoint_final.pt"
      ],
      "train_log": [
        {
          "stage": "answer_only",
          "local_step": 1,
          "step": 1,
          "loss": 2.8117666244506836,
          "trace_loss_active": false,
          "lr_scale": 1.0,
          "executor_loss": 2.8117666244506836,
          "standard_len2_direct_accuracy": NaN,
          "standard_len2_direct_target_mass": NaN,
          "standard_len2_executor_accuracy": 0.0,
          "standard_len2_executor_target_mass": 0.062022159496943154,
          "standard_len2_init_accuracy": 0.0,
          "standard_len2_op_accuracy": 0.4583333333333333,
          "standard_len2_arg_accuracy": 0.041666666666666664,
          "standard_len2_program_exact": 0.0,
          "standard_len4_direct_accuracy": NaN,
          "standard_len4_direct_target_mass": NaN,
          "standard_len4_executor_accuracy": 0.16666666666666666,
          "standard_len4_executor_target_mass": 0.057186623414357506,
          "standard_len4_init_accuracy": 0.16666666666666666,
          "standard_len4_op_accuracy": 0.2708333333333333,
          "standard_len4_arg_accuracy": 0.08333333333333333,
          "standard_len4_program_exact": 0.0,
          "paraphrase_len2_direct_accuracy": NaN,
          "paraphrase_len2_direct_target_mass": NaN,
          "paraphrase_len2_executor_accuracy": 0.08333333333333333,
          "paraphrase_len2_executor_target_mass": 0.05920091271400452,
          "paraphrase_len2_init_accuracy": 0.0,
          "paraphrase_len2_op_accuracy": 0.4166666666666667,
          "paraphrase_len2_arg_accuracy": 0.041666666666666664,
          "paraphrase_len2_program_exact": 0.0,
          "paraphrase_len4_direct_accuracy": NaN,
          "paraphrase_len4_direct_target_mass": NaN,
          "paraphrase_len4_executor_accuracy": 0.0,
          "paraphrase_len4_executor_target_mass": 0.05960758527119955,
          "paraphrase_len4_init_accuracy": 0.16666666666666666,
          "paraphrase_len4_op_accuracy": 0.375,
          "paraphrase_len4_arg_accuracy": 0.08333333333333333,
          "paraphrase_len4_program_exact": 0.0
        },
        {
          "stage": "answer_only",
          "local_step": 2,
          "step": 2,
          "loss": 2.8575143814086914,
          "trace_loss_active": false,
          "lr_scale": 1.0,
          "executor_loss": 2.8575143814086914,
          "standard_len2_direct_accuracy": NaN,
          "standard_len2_direct_target_mass": NaN,
          "standard_len2_executor_accuracy": 0.0,
          "standard_len2_executor_target_mass": 0.06177163124084473,
          "standard_len2_init_accuracy": 0.0,
          "standard_len2_op_accuracy": 0.4583333333333333,
          "standard_len2_arg_accuracy": 0.041666666666666664,
          "standard_len2_program_exact": 0.0,
          "standard_len4_direct_accuracy": NaN,
          "standard_len4_direct_target_mass": NaN,
          "standard_len4_executor_accuracy": 0.0,
          "standard_len4_executor_target_mass": 0.05730965733528137,
          "standard_len4_init_accuracy": 0.08333333333333333,
          "standard_len4_op_accuracy": 0.2708333333333333,
          "standard_len4_arg_accuracy": 0.125,
          "standard_len4_program_exact": 0.0,
          "paraphrase_len2_direct_accuracy": NaN,
          "paraphrase_len2_direct_target_mass": NaN,
          "paraphrase_len2_executor_accuracy": 0.08333333333333333,
          "paraphrase_len2_executor_target_mass": 0.05917522807916006,
          "paraphrase_len2_init_accuracy": 0.0,
          "paraphrase_len2_op_accuracy": 0.4166666666666667,
          "paraphrase_len2_arg_accuracy": 0.041666666666666664,
          "paraphrase_len2_program_exact": 0.0,
          "paraphrase_len4_direct_accuracy": NaN,
          "paraphrase_len4_direct_target_mass": NaN,
          "paraphrase_len4_executor_accuracy": 0.0,
          "paraphrase_len4_executor_target_mass": 0.0595710426568985,
          "paraphrase_len4_init_accuracy": 0.16666666666666666,
          "paraphrase_len4_op_accuracy": 0.3541666666666667,
          "paraphrase_len4_arg_accuracy": 0.08333333333333333,
          "paraphrase_len4_program_exact": 0.0
        },
        {
          "stage": "answer_only",
          "local_step": 4,
          "step": 4,
          "loss": 2.7949118614196777,
          "trace_loss_active": false,
          "lr_scale": 1.0,
          "executor_loss": 2.7949118614196777,
          "standard_len2_direct_accuracy": NaN,
          "standard_len2_direct_target_mass": NaN,
          "standard_len2_executor_accuracy": 0.0,
          "standard_len2_executor_target_mass": 0.06166746218999227,
          "standard_len2_init_accuracy": 0.0,
          "standard_len2_op_accuracy": 0.4166666666666667,
          "standard_len2_arg_accuracy": 0.041666666666666664,
          "standard_len2_program_exact": 0.0,
          "standard_len4_direct_accuracy": NaN,
          "standard_len4_direct_target_mass": NaN,
          "standard_len4_executor_accuracy": 0.0,
          "standard_len4_executor_target_mass": 0.05710078775882721,
          "standard_len4_init_accuracy": 0.08333333333333333,
          "standard_len4_op_accuracy": 0.2708333333333333,
          "standard_len4_arg_accuracy": 0.125,
          "standard_len4_program_exact": 0.0,
          "paraphrase_len2_direct_accuracy": NaN,
          "paraphrase_len2_direct_target_mass": NaN,
          "paraphrase_len2_executor_accuracy": 0.08333333333333333,
          "paraphrase_len2_executor_target_mass": 0.05924408137798309,
          "paraphrase_len2_init_accuracy": 0.0,
          "paraphrase_len2_op_accuracy": 0.4166666666666667,
          "paraphrase_len2_arg_accuracy": 0.041666666666666664,
          "paraphrase_len2_program_exact": 0.0,
          "paraphrase_len4_direct_accuracy": NaN,
          "paraphrase_len4_direct_target_mass": NaN,
          "paraphrase_len4_executor_accuracy": 0.0,
          "paraphrase_len4_executor_target_mass": 0.05968311925729116,
          "paraphrase_len4_init_accuracy": 0.16666666666666666,
          "paraphrase_len4_op_accuracy": 0.3125,
          "paraphrase_len4_arg_accuracy": 0.08333333333333333,
          "paraphrase_len4_program_exact": 0.0
        },
        {
          "stage": "answer_only",
          "local_step": 6,
          "step": 6,
          "loss": 2.861410140991211,
          "trace_loss_active": false,
          "lr_scale": 1.0,
          "executor_loss": 2.861410140991211,
          "standard_len2_direct_accuracy": NaN,
          "standard_len2_direct_target_mass": NaN,
          "standard_len2_executor_accuracy": 0.0,
          "standard_len2_executor_target_mass": 0.06135548154513041,
          "standard_len2_init_accuracy": 0.0,
          "standard_len2_op_accuracy": 0.375,
          "standard_len2_arg_accuracy": 0.041666666666666664,
          "standard_len2_program_exact": 0.0,
          "standard_len4_direct_accuracy": NaN,
          "standard_len4_direct_target_mass": NaN,
          "standard_len4_executor_accuracy": 0.0,
          "standard_len4_executor_target_mass": 0.057129114866256714,
          "standard_len4_init_accuracy": 0.08333333333333333,
          "standard_len4_op_accuracy": 0.25,
          "standard_len4_arg_accuracy": 0.125,
          "standard_len4_program_exact": 0.0,
          "paraphrase_len2_direct_accuracy": NaN,
          "paraphrase_len2_direct_target_mass": NaN,
          "paraphrase_len2_executor_accuracy": 0.08333333333333333,
          "paraphrase_len2_executor_target_mass": 0.059241016705830894,
          "paraphrase_len2_init_accuracy": 0.0,
          "paraphrase_len2_op_accuracy": 0.4166666666666667,
          "paraphrase_len2_arg_accuracy": 0.041666666666666664,
          "paraphrase_len2_program_exact": 0.0,
          "paraphrase_len4_direct_accuracy": NaN,
          "paraphrase_len4_direct_target_mass": NaN,
          "paraphrase_len4_executor_accuracy": 0.0,
          "paraphrase_len4_executor_target_mass": 0.05968991915384928,
          "paraphrase_len4_init_accuracy": 0.16666666666666666,
          "paraphrase_len4_op_accuracy": 0.3541666666666667,
          "paraphrase_len4_arg_accuracy": 0.08333333333333333,
          "paraphrase_len4_program_exact": 0.0
        },
        {
          "stage": "answer_only",
          "local_step": 8,
          "step": 8,
          "loss": 2.799020290374756,
          "trace_loss_active": false,
          "lr_scale": 1.0,
          "executor_loss": 2.799020290374756,
          "standard_len2_direct_accuracy": NaN,
          "standard_len2_direct_target_mass": NaN,
          "standard_len2_executor_accuracy": 0.08333333333333333,
          "standard_len2_executor_target_mass": 0.06090379257996877,
          "standard_len2_init_accuracy": 0.0,
          "standard_len2_op_accuracy": 0.3333333333333333,
          "standard_len2_arg_accuracy": 0.041666666666666664,
          "standard_len2_program_exact": 0.0,
          "standard_len4_direct_accuracy": NaN,
          "standard_len4_direct_target_mass": NaN,
          "standard_len4_executor_accuracy": 0.08333333333333333,
          "standard_len4_executor_target_mass": 0.057366758584976196,
          "standard_len4_init_accuracy": 0.08333333333333333,
          "standard_len4_op_accuracy": 0.25,
          "standard_len4_arg_accuracy": 0.125,
          "standard_len4_program_exact": 0.0,
          "paraphrase_len2_direct_accuracy": NaN,
          "paraphrase_len2_direct_target_mass": NaN,
          "paraphrase_len2_executor_accuracy": 0.08333333333333333,
          "paraphrase_len2_executor_target_mass": 0.05920960505803426,
          "paraphrase_len2_init_accuracy": 0.0,
          "paraphrase_len2_op_accuracy": 0.375,
          "paraphrase_len2_arg_accuracy": 0.041666666666666664,
          "paraphrase_len2_program_exact": 0.0,
          "paraphrase_len4_direct_accuracy": NaN,
          "paraphrase_len4_direct_target_mass": NaN,
          "paraphrase_len4_executor_accuracy": 0.0,
          "paraphrase_len4_executor_target_mass": 0.05961151917775472,
          "paraphrase_len4_init_accuracy": 0.16666666666666666,
          "paraphrase_len4_op_accuracy": 0.3125,
          "paraphrase_len4_arg_accuracy": 0.08333333333333333,
          "paraphrase_len4_program_exact": 0.0
        }
      ],
      "final_metrics": {
        "standard_len2": {
          "variant": "compiler_answer_only",
          "n": 12.0,
          "direct_accuracy": NaN,
          "direct_target_mass": NaN,
          "executor_accuracy": 0.08333333333333333,
          "executor_target_mass": 0.06090379257996877,
          "init_accuracy": 0.0,
          "op_accuracy": 0.3333333333333333,
          "arg_accuracy": 0.041666666666666664,
          "program_exact": 0.0
        },
        "standard_len4": {
          "variant": "compiler_answer_only",
          "n": 12.0,
          "direct_accuracy": NaN,
          "direct_target_mass": NaN,
          "executor_accuracy": 0.08333333333333333,
          "executor_target_mass": 0.057366758584976196,
          "init_accuracy": 0.08333333333333333,
          "op_accuracy": 0.25,
          "arg_accuracy": 0.125,
          "program_exact": 0.0
        },
        "paraphrase_len2": {
          "variant": "compiler_answer_only",
          "n": 12.0,
          "direct_accuracy": NaN,
          "direct_target_mass": NaN,
          "executor_accuracy": 0.08333333333333333,
          "executor_target_mass": 0.05920960505803426,
          "init_accuracy": 0.0,
          "op_accuracy": 0.375,
          "arg_accuracy": 0.041666666666666664,
          "program_exact": 0.0
        },
        "paraphrase_len4": {
          "variant": "compiler_answer_only",
          "n": 12.0,
          "direct_accuracy": NaN,
          "direct_target_mass": NaN,
          "executor_accuracy": 0.0,
          "executor_target_mass": 0.05961151917775472,
          "init_accuracy": 0.16666666666666666,
          "op_accuracy": 0.3125,
          "arg_accuracy": 0.08333333333333333,
          "program_exact": 0.0
        }
      }
    },
    "compiler_trace_then_answer": {
      "variant": "compiler_trace_then_answer",
      "train_seconds": 0.8049569129943848,
      "checkpoints": [
        "large_artifacts/qwen_span_free_compiler/checkpoints/smoke_tiny/compiler_trace_then_answer/checkpoint_final.pt"
      ],
      "train_log": [
        {
          "stage": "trace_bootstrap",
          "local_step": 1,
          "step": 1,
          "loss": 9.505542755126953,
          "trace_loss_active": true,
          "lr_scale": 1.0,
          "executor_loss": 2.8132424354553223,
          "init_loss": 2.703650712966919,
          "op_loss": 1.1107409000396729,
          "arg_loss": 2.87790846824646,
          "standard_len2_direct_accuracy": NaN,
          "standard_len2_direct_target_mass": NaN,
          "standard_len2_executor_accuracy": 0.0,
          "standard_len2_executor_target_mass": 0.06068208316961924,
          "standard_len2_init_accuracy": 0.0,
          "standard_len2_op_accuracy": 0.2916666666666667,
          "standard_len2_arg_accuracy": 0.0,
          "standard_len2_program_exact": 0.0,
          "standard_len4_direct_accuracy": NaN,
          "standard_len4_direct_target_mass": NaN,
          "standard_len4_executor_accuracy": 0.08333333333333333,
          "standard_len4_executor_target_mass": 0.05774883429209391,
          "standard_len4_init_accuracy": 0.08333333333333333,
          "standard_len4_op_accuracy": 0.3541666666666667,
          "standard_len4_arg_accuracy": 0.08333333333333333,
          "standard_len4_program_exact": 0.0,
          "paraphrase_len2_direct_accuracy": NaN,
          "paraphrase_len2_direct_target_mass": NaN,
          "paraphrase_len2_executor_accuracy": 0.16666666666666666,
          "paraphrase_len2_executor_target_mass": 0.059431999921798706,
          "paraphrase_len2_init_accuracy": 0.08333333333333333,
          "paraphrase_len2_op_accuracy": 0.25,
          "paraphrase_len2_arg_accuracy": 0.0,
          "paraphrase_len2_program_exact": 0.0,
          "paraphrase_len4_direct_accuracy": NaN,
          "paraphrase_len4_direct_target_mass": NaN,
          "paraphrase_len4_executor_accuracy": 0.0,
          "paraphrase_len4_executor_target_mass": 0.0589027206103007,
          "paraphrase_len4_init_accuracy": 0.0,
          "paraphrase_len4_op_accuracy": 0.3125,
          "paraphrase_len4_arg_accuracy": 0.0625,
          "paraphrase_len4_program_exact": 0.0
        },
        {
          "stage": "trace_bootstrap",
          "local_step": 2,
          "step": 2,
          "loss": 9.790613174438477,
          "trace_loss_active": true,
          "lr_scale": 1.0,
          "executor_loss": 2.846224069595337,
          "init_loss": 2.898372173309326,
          "op_loss": 1.2120521068572998,
          "arg_loss": 2.8339648246765137,
          "standard_len2_direct_accuracy": NaN,
          "standard_len2_direct_target_mass": NaN,
          "standard_len2_executor_accuracy": 0.0,
          "standard_len2_executor_target_mass": 0.06074431538581848,
          "standard_len2_init_accuracy": 0.0,
          "standard_len2_op_accuracy": 0.3333333333333333,
          "standard_len2_arg_accuracy": 0.0,
          "standard_len2_program_exact": 0.0,
          "standard_len4_direct_accuracy": NaN,
          "standard_len4_direct_target_mass": NaN,
          "standard_len4_executor_accuracy": 0.08333333333333333,
          "standard_len4_executor_target_mass": 0.057691325743993126,
          "standard_len4_init_accuracy": 0.08333333333333333,
          "standard_len4_op_accuracy": 0.3333333333333333,
          "standard_len4_arg_accuracy": 0.0625,
          "standard_len4_program_exact": 0.0,
          "paraphrase_len2_direct_accuracy": NaN,
          "paraphrase_len2_direct_target_mass": NaN,
          "paraphrase_len2_executor_accuracy": 0.08333333333333333,
          "paraphrase_len2_executor_target_mass": 0.05942369004090627,
          "paraphrase_len2_init_accuracy": 0.08333333333333333,
          "paraphrase_len2_op_accuracy": 0.20833333333333334,
          "paraphrase_len2_arg_accuracy": 0.0,
          "paraphrase_len2_program_exact": 0.0,
          "paraphrase_len4_direct_accuracy": NaN,
          "paraphrase_len4_direct_target_mass": NaN,
          "paraphrase_len4_executor_accuracy": 0.08333333333333333,
          "paraphrase_len4_executor_target_mass": 0.058916514118512474,
          "paraphrase_len4_init_accuracy": 0.0,
          "paraphrase_len4_op_accuracy": 0.375,
          "paraphrase_len4_arg_accuracy": 0.0625,
          "paraphrase_len4_program_exact": 0.0
        },
        {
          "stage": "trace_bootstrap",
          "local_step": 4,
          "step": 4,
          "loss": 9.437228202819824,
          "trace_loss_active": true,
          "lr_scale": 1.0,
          "executor_loss": 2.846967935562134,
          "init_loss": 2.7418174743652344,
          "op_loss": 1.081488847732544,
          "arg_loss": 2.766954183578491,
          "standard_len2_direct_accuracy": NaN,
          "standard_len2_direct_target_mass": NaN,
          "standard_len2_executor_accuracy": 0.0,
          "standard_len2_executor_target_mass": 0.06075367331504822,
          "standard_len2_init_accuracy": 0.0,
          "standard_len2_op_accuracy": 0.25,
          "standard_len2_arg_accuracy": 0.0,
          "standard_len2_program_exact": 0.0,
          "standard_len4_direct_accuracy": NaN,
          "standard_len4_direct_target_mass": NaN,
          "standard_len4_executor_accuracy": 0.16666666666666666,
          "standard_len4_executor_target_mass": 0.0576278418302536,
          "standard_len4_init_accuracy": 0.08333333333333333,
          "standard_len4_op_accuracy": 0.3125,
          "standard_len4_arg_accuracy": 0.041666666666666664,
          "standard_len4_program_exact": 0.0,
          "paraphrase_len2_direct_accuracy": NaN,
          "paraphrase_len2_direct_target_mass": NaN,
          "paraphrase_len2_executor_accuracy": 0.08333333333333333,
          "paraphrase_len2_executor_target_mass": 0.05937573313713074,
          "paraphrase_len2_init_accuracy": 0.08333333333333333,
          "paraphrase_len2_op_accuracy": 0.3333333333333333,
          "paraphrase_len2_arg_accuracy": 0.041666666666666664,
          "paraphrase_len2_program_exact": 0.0,
          "paraphrase_len4_direct_accuracy": NaN,
          "paraphrase_len4_direct_target_mass": NaN,
          "paraphrase_len4_executor_accuracy": 0.08333333333333333,
          "paraphrase_len4_executor_target_mass": 0.05894559621810913,
          "paraphrase_len4_init_accuracy": 0.0,
          "paraphrase_len4_op_accuracy": 0.3541666666666667,
          "paraphrase_len4_arg_accuracy": 0.0625,
          "paraphrase_len4_program_exact": 0.0
        },
        {
          "stage": "answer_retention",
          "local_step": 1,
          "step": 5,
          "loss": 2.851250648498535,
          "trace_loss_active": false,
          "lr_scale": 1.0,
          "executor_loss": 2.851250648498535,
          "standard_len2_direct_accuracy": NaN,
          "standard_len2_direct_target_mass": NaN,
          "standard_len2_executor_accuracy": 0.0,
          "standard_len2_executor_target_mass": 0.060712004701296486,
          "standard_len2_init_accuracy": 0.0,
          "standard_len2_op_accuracy": 0.2916666666666667,
          "standard_len2_arg_accuracy": 0.0,
          "standard_len2_program_exact": 0.0,
          "standard_len4_direct_accuracy": NaN,
          "standard_len4_direct_target_mass": NaN,
          "standard_len4_executor_accuracy": 0.16666666666666666,
          "standard_len4_executor_target_mass": 0.05763504902521769,
          "standard_len4_init_accuracy": 0.08333333333333333,
          "standard_len4_op_accuracy": 0.3125,
          "standard_len4_arg_accuracy": 0.0625,
          "standard_len4_program_exact": 0.0,
          "paraphrase_len2_direct_accuracy": NaN,
          "paraphrase_len2_direct_target_mass": NaN,
          "paraphrase_len2_executor_accuracy": 0.08333333333333333,
          "paraphrase_len2_executor_target_mass": 0.05934560795625051,
          "paraphrase_len2_init_accuracy": 0.08333333333333333,
          "paraphrase_len2_op_accuracy": 0.375,
          "paraphrase_len2_arg_accuracy": 0.041666666666666664,
          "paraphrase_len2_program_exact": 0.0,
          "paraphrase_len4_direct_accuracy": NaN,
          "paraphrase_len4_direct_target_mass": NaN,
          "paraphrase_len4_executor_accuracy": 0.0,
          "paraphrase_len4_executor_target_mass": 0.058950444062550865,
          "paraphrase_len4_init_accuracy": 0.0,
          "paraphrase_len4_op_accuracy": 0.4166666666666667,
          "paraphrase_len4_arg_accuracy": 0.0625,
          "paraphrase_len4_program_exact": 0.0
        },
        {
          "stage": "answer_retention",
          "local_step": 2,
          "step": 6,
          "loss": 2.7808358669281006,
          "trace_loss_active": false,
          "lr_scale": 1.0,
          "executor_loss": 2.7808358669281006,
          "standard_len2_direct_accuracy": NaN,
          "standard_len2_direct_target_mass": NaN,
          "standard_len2_executor_accuracy": 0.08333333333333333,
          "standard_len2_executor_target_mass": 0.06072754661242167,
          "standard_len2_init_accuracy": 0.0,
          "standard_len2_op_accuracy": 0.2916666666666667,
          "standard_len2_arg_accuracy": 0.0,
          "standard_len2_program_exact": 0.0,
          "standard_len4_direct_accuracy": NaN,
          "standard_len4_direct_target_mass": NaN,
          "standard_len4_executor_accuracy": 0.0,
          "standard_len4_executor_target_mass": 0.05760765075683594,
          "standard_len4_init_accuracy": 0.08333333333333333,
          "standard_len4_op_accuracy": 0.2708333333333333,
          "standard_len4_arg_accuracy": 0.041666666666666664,
          "standard_len4_program_exact": 0.0,
          "paraphrase_len2_direct_accuracy": NaN,
          "paraphrase_len2_direct_target_mass": NaN,
          "paraphrase_len2_executor_accuracy": 0.08333333333333333,
          "paraphrase_len2_executor_target_mass": 0.05933367212613424,
          "paraphrase_len2_init_accuracy": 0.08333333333333333,
          "paraphrase_len2_op_accuracy": 0.375,
          "paraphrase_len2_arg_accuracy": 0.08333333333333333,
          "paraphrase_len2_program_exact": 0.0,
          "paraphrase_len4_direct_accuracy": NaN,
          "paraphrase_len4_direct_target_mass": NaN,
          "paraphrase_len4_executor_accuracy": 0.0,
          "paraphrase_len4_executor_target_mass": 0.058960373202959694,
          "paraphrase_len4_init_accuracy": 0.0,
          "paraphrase_len4_op_accuracy": 0.4375,
          "paraphrase_len4_arg_accuracy": 0.0625,
          "paraphrase_len4_program_exact": 0.0
        },
        {
          "stage": "answer_retention",
          "local_step": 4,
          "step": 8,
          "loss": 2.8143978118896484,
          "trace_loss_active": false,
          "lr_scale": 1.0,
          "executor_loss": 2.8143978118896484,
          "standard_len2_direct_accuracy": NaN,
          "standard_len2_direct_target_mass": NaN,
          "standard_len2_executor_accuracy": 0.08333333333333333,
          "standard_len2_executor_target_mass": 0.06084983547528585,
          "standard_len2_init_accuracy": 0.0,
          "standard_len2_op_accuracy": 0.25,
          "standard_len2_arg_accuracy": 0.0,
          "standard_len2_program_exact": 0.0,
          "standard_len4_direct_accuracy": NaN,
          "standard_len4_direct_target_mass": NaN,
          "standard_len4_executor_accuracy": 0.0,
          "standard_len4_executor_target_mass": 0.05750989417235056,
          "standard_len4_init_accuracy": 0.08333333333333333,
          "standard_len4_op_accuracy": 0.22916666666666666,
          "standard_len4_arg_accuracy": 0.041666666666666664,
          "standard_len4_program_exact": 0.0,
          "paraphrase_len2_direct_accuracy": NaN,
          "paraphrase_len2_direct_target_mass": NaN,
          "paraphrase_len2_executor_accuracy": 0.08333333333333333,
          "paraphrase_len2_executor_target_mass": 0.05933640400568644,
          "paraphrase_len2_init_accuracy": 0.08333333333333333,
          "paraphrase_len2_op_accuracy": 0.4166666666666667,
          "paraphrase_len2_arg_accuracy": 0.125,
          "paraphrase_len2_program_exact": 0.0,
          "paraphrase_len4_direct_accuracy": NaN,
          "paraphrase_len4_direct_target_mass": NaN,
          "paraphrase_len4_executor_accuracy": 0.0,
          "paraphrase_len4_executor_target_mass": 0.05898126463095347,
          "paraphrase_len4_init_accuracy": 0.0,
          "paraphrase_len4_op_accuracy": 0.5,
          "paraphrase_len4_arg_accuracy": 0.08333333333333333,
          "paraphrase_len4_program_exact": 0.0
        }
      ],
      "final_metrics": {
        "standard_len2": {
          "variant": "compiler_trace_then_answer",
          "n": 12.0,
          "direct_accuracy": NaN,
          "direct_target_mass": NaN,
          "executor_accuracy": 0.08333333333333333,
          "executor_target_mass": 0.06084983547528585,
          "init_accuracy": 0.0,
          "op_accuracy": 0.25,
          "arg_accuracy": 0.0,
          "program_exact": 0.0
        },
        "standard_len4": {
          "variant": "compiler_trace_then_answer",
          "n": 12.0,
          "direct_accuracy": NaN,
          "direct_target_mass": NaN,
          "executor_accuracy": 0.0,
          "executor_target_mass": 0.05750989417235056,
          "init_accuracy": 0.08333333333333333,
          "op_accuracy": 0.22916666666666666,
          "arg_accuracy": 0.041666666666666664,
          "program_exact": 0.0
        },
        "paraphrase_len2": {
          "variant": "compiler_trace_then_answer",
          "n": 12.0,
          "direct_accuracy": NaN,
          "direct_target_mass": NaN,
          "executor_accuracy": 0.08333333333333333,
          "executor_target_mass": 0.05933640400568644,
          "init_accuracy": 0.08333333333333333,
          "op_accuracy": 0.4166666666666667,
          "arg_accuracy": 0.125,
          "program_exact": 0.0
        },
        "paraphrase_len4": {
          "variant": "compiler_trace_then_answer",
          "n": 12.0,
          "direct_accuracy": NaN,
          "direct_target_mass": NaN,
          "executor_accuracy": 0.0,
          "executor_target_mass": 0.05898126463095347,
          "init_accuracy": 0.0,
          "op_accuracy": 0.5,
          "arg_accuracy": 0.08333333333333333,
          "program_exact": 0.0
        }
      }
    }
  }
}