{
  "args": {
    "model_id": "Qwen/Qwen3-4B",
    "load_in_4bit": true,
    "torch_dtype": "bf16",
    "device_map": "auto",
    "use_lora": true,
    "lora_r": 4,
    "lora_alpha": 8,
    "lora_dropout": 0.05,
    "lora_target_modules": "all-linear",
    "gradient_checkpointing": true,
    "optimizer": "paged_adamw_8bit",
    "modulus": 97,
    "max_steps": 24,
    "beam_count": 4,
    "train_min_len": 1,
    "train_max_len": 12,
    "curriculum_stages": "short:1:4:30,medium:1:8:30,train:1:12:30,long:8:24:30",
    "eval_lengths": "12,24",
    "train_template_mode": "mixed",
    "paired_train": true,
    "paired_batches": true,
    "paired_template_modes": "standard,paraphrase",
    "pair_size": 2,
    "eval_template_modes": "standard,paraphrase",
    "paired_eval": true,
    "paired_eval_template_modes": "standard,paraphrase",
    "train_size": 64,
    "eval_size": 32,
    "train_batch_size": 2,
    "eval_batch_size": 2,
    "eval_every": 60,
    "stage_eval_every": 30,
    "lr": 0.0002,
    "weight_decay": 0.0,
    "grad_clip": 1.0,
    "register_width": 256,
    "register_layers": 1,
    "register_heads": 4,
    "register_dropout": 0.05,
    "trace_loss_weight": 1.0,
    "init_trace_loss_weight": 4.0,
    "op_trace_loss_weight": 1.0,
    "arg_trace_loss_weight": 4.0,
    "executor_loss_weight": 1.0,
    "state_loss_weight": 1.0,
    "selector_loss_weight": 0.2,
    "diversity_loss_weight": 0.03,
    "pair_set_consistency_loss_weight": 0.05,
    "set_loss_tau": 0.5,
    "pair_set_tau": 0.5,
    "max_length": 3072,
    "seed": 11,
    "output_dir": "experiments/qwen_latent_beam_program_compiler/runs/pilot_beam4_s120",
    "checkpoint_dir": "large_artifacts/qwen_latent_beam_program_compiler/checkpoints/pilot_beam4_s120"
  },
  "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,
    "use_lora": true,
    "lora_r": 4,
    "gpu_name": "NVIDIA RTX 6000 Ada Generation",
    "gpu_vram_gb": 47.383
  },
  "hidden_dim": 2560,
  "beam_count": 4,
  "register_count": 49,
  "dataset": {
    "train_size": 64,
    "curriculum_stages": [
      {
        "name": "short",
        "min_len": 1,
        "max_len": 4,
        "steps": 30
      },
      {
        "name": "medium",
        "min_len": 1,
        "max_len": 8,
        "steps": 30
      },
      {
        "name": "train",
        "min_len": 1,
        "max_len": 12,
        "steps": 30
      },
      {
        "name": "long",
        "min_len": 8,
        "max_len": 24,
        "steps": 30
      }
    ],
    "eval_size": 32,
    "eval_lengths": [
      12,
      24
    ],
    "train_template_mode": "mixed",
    "paired_train": true,
    "paired_template_modes": [
      "standard",
      "paraphrase"
    ],
    "eval_template_modes": [
      "standard",
      "paraphrase"
    ],
    "paired_eval": true,
    "paired_eval_template_modes": [
      "standard",
      "paraphrase"
    ]
  },
  "train_seconds": 1611.4617655277252,
  "checkpoints": [
    "large_artifacts/qwen_latent_beam_program_compiler/checkpoints/pilot_beam4_s120/adapter",
    "large_artifacts/qwen_latent_beam_program_compiler/checkpoints/pilot_beam4_s120/heads.pt"
  ],
  "train_log": [
    {
      "stage": "short",
      "local_step": 1,
      "step": 1,
      "loss": 45.65517807006836,
      "trace_set_loss": 36.94873046875,
      "executor_loss": 4.576757431030273,
      "state_set_loss": 3.8836100101470947,
      "selector_loss": 1.4021222591400146,
      "diversity_loss": 0.010319283232092857,
      "pair_set_loss": -0.6931312084197998,
      "selector_matches_best_trace": 0.0,
      "best_soft_answer_mass": 0.010294589214026928,
      "mean_soft_answer_mass": 0.010288202203810215,
      "best_target_prob_beam_fraction": 0.0,
      "standard_len12_n": 32.0,
      "standard_len12_beam_count": 4.0,
      "standard_len12_beam0_accuracy": 0.0,
      "standard_len12_selected_accuracy": 0.0,
      "standard_len12_prior_accuracy": 0.0,
      "standard_len12_oracle_accuracy": 0.0,
      "standard_len12_beam0_program_exact": 0.0,
      "standard_len12_selected_program_exact": 0.0,
      "standard_len12_prior_program_exact": 0.0,
      "standard_len12_oracle_program_exact": 0.0,
      "standard_len12_selected_state_prefix_fraction": 0.0,
      "standard_len12_oracle_state_prefix_fraction": 0.0,
      "standard_len12_avg_distinct_answers": 2.78125,
      "standard_len12_selector_entropy": 1.3862679302692413,
      "standard_len12_selected_oracle_gap_recovered": NaN,
      "standard_len24_n": 32.0,
      "standard_len24_beam_count": 4.0,
      "standard_len24_beam0_accuracy": 0.0,
      "standard_len24_selected_accuracy": 0.0,
      "standard_len24_prior_accuracy": 0.0,
      "standard_len24_oracle_accuracy": 0.0,
      "standard_len24_beam0_program_exact": 0.0,
      "standard_len24_selected_program_exact": 0.0,
      "standard_len24_prior_program_exact": 0.0,
      "standard_len24_oracle_program_exact": 0.0,
      "standard_len24_selected_state_prefix_fraction": 0.0,
      "standard_len24_oracle_state_prefix_fraction": 0.0,
      "standard_len24_avg_distinct_answers": 4.0,
      "standard_len24_selector_entropy": 1.3862828686833382,
      "standard_len24_selected_oracle_gap_recovered": NaN,
      "paraphrase_len12_n": 32.0,
      "paraphrase_len12_beam_count": 4.0,
      "paraphrase_len12_beam0_accuracy": 0.03125,
      "paraphrase_len12_selected_accuracy": 0.0,
      "paraphrase_len12_prior_accuracy": 0.03125,
      "paraphrase_len12_oracle_accuracy": 0.0625,
      "paraphrase_len12_beam0_program_exact": 0.0,
      "paraphrase_len12_selected_program_exact": 0.0,
      "paraphrase_len12_prior_program_exact": 0.0,
      "paraphrase_len12_oracle_program_exact": 0.0,
      "paraphrase_len12_selected_state_prefix_fraction": 0.0,
      "paraphrase_len12_oracle_state_prefix_fraction": 0.0,
      "paraphrase_len12_avg_distinct_answers": 2.84375,
      "paraphrase_len12_selector_entropy": 1.3862804546952248,
      "paraphrase_len12_selected_oracle_gap_recovered": -1.0,
      "paraphrase_len24_n": 32.0,
      "paraphrase_len24_beam_count": 4.0,
      "paraphrase_len24_beam0_accuracy": 0.0,
      "paraphrase_len24_selected_accuracy": 0.0625,
      "paraphrase_len24_prior_accuracy": 0.0,
      "paraphrase_len24_oracle_accuracy": 0.09375,
      "paraphrase_len24_beam0_program_exact": 0.0,
      "paraphrase_len24_selected_program_exact": 0.0,
      "paraphrase_len24_prior_program_exact": 0.0,
      "paraphrase_len24_oracle_program_exact": 0.0,
      "paraphrase_len24_selected_state_prefix_fraction": 0.0,
      "paraphrase_len24_oracle_state_prefix_fraction": 0.0,
      "paraphrase_len24_avg_distinct_answers": 3.875,
      "paraphrase_len24_selector_entropy": 1.386277548968792,
      "paraphrase_len24_selected_oracle_gap_recovered": 0.6666666666666666,
      "paired_len12_n": 64.0,
      "paired_len12_beam_count": 4.0,
      "paired_len12_beam0_accuracy": 0.0,
      "paired_len12_selected_accuracy": 0.0,
      "paired_len12_prior_accuracy": 0.0,
      "paired_len12_oracle_accuracy": 0.03125,
      "paired_len12_beam0_program_exact": 0.0,
      "paired_len12_selected_program_exact": 0.0,
      "paired_len12_prior_program_exact": 0.0,
      "paired_len12_oracle_program_exact": 0.0,
      "paired_len12_selected_state_prefix_fraction": 0.0026041666666666665,
      "paired_len12_oracle_state_prefix_fraction": 0.0026041666666666665,
      "paired_len12_avg_distinct_answers": 2.90625,
      "paired_len12_selector_entropy": 1.3862708732485771,
      "paired_len12_selected_oracle_gap_recovered": 0.0,
      "paired_len12_selected_pair_answer_consistency": 0.0625,
      "paired_len12_selected_pair_both_correct": 0.0,
      "paired_len12_oracle_pair_both_correct": 0.0,
      "paired_len12_selected_pair_state_consistency": 0.03125,
      "paired_len12_oracle_pair_state_consistency": 0.6875,
      "paired_len24_n": 64.0,
      "paired_len24_beam_count": 4.0,
      "paired_len24_beam0_accuracy": 0.0625,
      "paired_len24_selected_accuracy": 0.0,
      "paired_len24_prior_accuracy": 0.0625,
      "paired_len24_oracle_accuracy": 0.09375,
      "paired_len24_beam0_program_exact": 0.0,
      "paired_len24_selected_program_exact": 0.0,
      "paired_len24_prior_program_exact": 0.0,
      "paired_len24_oracle_program_exact": 0.0,
      "paired_len24_selected_state_prefix_fraction": 0.0013020833333333333,
      "paired_len24_oracle_state_prefix_fraction": 0.0013020833333333333,
      "paired_len24_avg_distinct_answers": 3.9375,
      "paired_len24_selector_entropy": 1.3862790130078793,
      "paired_len24_selected_oracle_gap_recovered": -2.0,
      "paired_len24_selected_pair_answer_consistency": 0.0,
      "paired_len24_selected_pair_both_correct": 0.0,
      "paired_len24_oracle_pair_both_correct": 0.0625,
      "paired_len24_selected_pair_state_consistency": 0.0,
      "paired_len24_oracle_pair_state_consistency": 0.6875
    },
    {
      "stage": "short",
      "local_step": 30,
      "step": 30,
      "loss": 44.62309265136719,
      "trace_set_loss": 35.91956329345703,
      "executor_loss": 4.577186584472656,
      "state_set_loss": 3.881450653076172,
      "selector_loss": 1.3962054252624512,
      "diversity_loss": 0.010312643833458424,
      "pair_set_loss": -0.6931463479995728,
      "selector_matches_best_trace": 0.0,
      "best_soft_answer_mass": 0.01028402615338564,
      "mean_soft_answer_mass": 0.010283788666129112,
      "best_target_prob_beam_fraction": 1.0,
      "standard_len12_n": 32.0,
      "standard_len12_beam_count": 4.0,
      "standard_len12_beam0_accuracy": 0.0,
      "standard_len12_selected_accuracy": 0.0,
      "standard_len12_prior_accuracy": 0.0,
      "standard_len12_oracle_accuracy": 0.0,
      "standard_len12_beam0_program_exact": 0.0,
      "standard_len12_selected_program_exact": 0.0,
      "standard_len12_prior_program_exact": 0.0,
      "standard_len12_oracle_program_exact": 0.0,
      "standard_len12_selected_state_prefix_fraction": 0.0026041666666666665,
      "standard_len12_oracle_state_prefix_fraction": 0.0,
      "standard_len12_avg_distinct_answers": 2.0,
      "standard_len12_selector_entropy": 1.3860917836427689,
      "standard_len12_selected_oracle_gap_recovered": NaN,
      "standard_len24_n": 32.0,
      "standard_len24_beam_count": 4.0,
      "standard_len24_beam0_accuracy": 0.0,
      "standard_len24_selected_accuracy": 0.0,
      "standard_len24_prior_accuracy": 0.0,
      "standard_len24_oracle_accuracy": 0.0,
      "standard_len24_beam0_program_exact": 0.0,
      "standard_len24_selected_program_exact": 0.0,
      "standard_len24_prior_program_exact": 0.0,
      "standard_len24_oracle_program_exact": 0.0,
      "standard_len24_selected_state_prefix_fraction": 0.0026041666666666665,
      "standard_len24_oracle_state_prefix_fraction": 0.0013020833333333333,
      "standard_len24_avg_distinct_answers": 2.0,
      "standard_len24_selector_entropy": 1.3860779032111168,
      "standard_len24_selected_oracle_gap_recovered": NaN,
      "paraphrase_len12_n": 32.0,
      "paraphrase_len12_beam_count": 4.0,
      "paraphrase_len12_beam0_accuracy": 0.03125,
      "paraphrase_len12_selected_accuracy": 0.03125,
      "paraphrase_len12_prior_accuracy": 0.0,
      "paraphrase_len12_oracle_accuracy": 0.03125,
      "paraphrase_len12_beam0_program_exact": 0.0,
      "paraphrase_len12_selected_program_exact": 0.0,
      "paraphrase_len12_prior_program_exact": 0.0,
      "paraphrase_len12_oracle_program_exact": 0.0,
      "paraphrase_len12_selected_state_prefix_fraction": 0.0,
      "paraphrase_len12_oracle_state_prefix_fraction": 0.0,
      "paraphrase_len12_avg_distinct_answers": 2.0,
      "paraphrase_len12_selector_entropy": 1.3861039504408836,
      "paraphrase_len12_selected_oracle_gap_recovered": NaN,
      "paraphrase_len24_n": 32.0,
      "paraphrase_len24_beam_count": 4.0,
      "paraphrase_len24_beam0_accuracy": 0.03125,
      "paraphrase_len24_selected_accuracy": 0.03125,
      "paraphrase_len24_prior_accuracy": 0.0625,
      "paraphrase_len24_oracle_accuracy": 0.09375,
      "paraphrase_len24_beam0_program_exact": 0.0,
      "paraphrase_len24_selected_program_exact": 0.0,
      "paraphrase_len24_prior_program_exact": 0.0,
      "paraphrase_len24_oracle_program_exact": 0.0,
      "paraphrase_len24_selected_state_prefix_fraction": 0.0,
      "paraphrase_len24_oracle_state_prefix_fraction": 0.0013020833333333333,
      "paraphrase_len24_avg_distinct_answers": 2.0,
      "paraphrase_len24_selector_entropy": 1.3860964626073837,
      "paraphrase_len24_selected_oracle_gap_recovered": 0.0,
      "paired_len12_n": 64.0,
      "paired_len12_beam_count": 4.0,
      "paired_len12_beam0_accuracy": 0.0,
      "paired_len12_selected_accuracy": 0.0,
      "paired_len12_prior_accuracy": 0.0,
      "paired_len12_oracle_accuracy": 0.0,
      "paired_len12_beam0_program_exact": 0.0,
      "paired_len12_selected_program_exact": 0.0,
      "paired_len12_prior_program_exact": 0.0,
      "paired_len12_oracle_program_exact": 0.0,
      "paired_len12_selected_state_prefix_fraction": 0.0026041666666666665,
      "paired_len12_oracle_state_prefix_fraction": 0.0,
      "paired_len12_avg_distinct_answers": 2.0,
      "paired_len12_selector_entropy": 1.3860961832106113,
      "paired_len12_selected_oracle_gap_recovered": NaN,
      "paired_len12_selected_pair_answer_consistency": 1.0,
      "paired_len12_selected_pair_both_correct": 0.0,
      "paired_len12_oracle_pair_both_correct": 0.0,
      "paired_len12_selected_pair_state_consistency": 1.0,
      "paired_len12_oracle_pair_state_consistency": 1.0,
      "paired_len24_n": 64.0,
      "paired_len24_beam_count": 4.0,
      "paired_len24_beam0_accuracy": 0.03125,
      "paired_len24_selected_accuracy": 0.03125,
      "paired_len24_prior_accuracy": 0.0,
      "paired_len24_oracle_accuracy": 0.03125,
      "paired_len24_beam0_program_exact": 0.0,
      "paired_len24_selected_program_exact": 0.0,
      "paired_len24_prior_program_exact": 0.0,
      "paired_len24_oracle_program_exact": 0.0,
      "paired_len24_selected_state_prefix_fraction": 0.0,
      "paired_len24_oracle_state_prefix_fraction": 0.0,
      "paired_len24_avg_distinct_answers": 2.0,
      "paired_len24_selector_entropy": 1.3860919550061226,
      "paired_len24_selected_oracle_gap_recovered": NaN,
      "paired_len24_selected_pair_answer_consistency": 1.0,
      "paired_len24_selected_pair_both_correct": 0.03125,
      "paired_len24_oracle_pair_both_correct": 0.03125,
      "paired_len24_selected_pair_state_consistency": 1.0,
      "paired_len24_oracle_pair_state_consistency": 0.96875
    },
    {
      "stage": "medium",
      "local_step": 1,
      "step": 31,
      "loss": 44.563236236572266,
      "trace_set_loss": 35.88201904296875,
      "executor_loss": 4.576687812805176,
      "state_set_loss": 3.867936134338379,
      "selector_loss": 1.354710578918457,
      "diversity_loss": 0.010312441736459732,
      "pair_set_loss": -0.693146824836731,
      "selector_matches_best_trace": 1.0,
      "best_soft_answer_mass": 0.010289555415511131,
      "mean_soft_answer_mass": 0.010288916528224945,
      "best_target_prob_beam_fraction": 0.0,
      "standard_len12_n": 32.0,
      "standard_len12_beam_count": 4.0,
      "standard_len12_beam0_accuracy": 0.0,
      "standard_len12_selected_accuracy": 0.0,
      "standard_len12_prior_accuracy": 0.0,
      "standard_len12_oracle_accuracy": 0.0,
      "standard_len12_beam0_program_exact": 0.0,
      "standard_len12_selected_program_exact": 0.0,
      "standard_len12_prior_program_exact": 0.0,
      "standard_len12_oracle_program_exact": 0.0,
      "standard_len12_selected_state_prefix_fraction": 0.0026041666666666665,
      "standard_len12_oracle_state_prefix_fraction": 0.0,
      "standard_len12_avg_distinct_answers": 2.0,
      "standard_len12_selector_entropy": 1.3858995139598846,
      "standard_len12_selected_oracle_gap_recovered": NaN,
      "standard_len24_n": 32.0,
      "standard_len24_beam_count": 4.0,
      "standard_len24_beam0_accuracy": 0.0,
      "standard_len24_selected_accuracy": 0.0,
      "standard_len24_prior_accuracy": 0.0,
      "standard_len24_oracle_accuracy": 0.0,
      "standard_len24_beam0_program_exact": 0.0,
      "standard_len24_selected_program_exact": 0.0,
      "standard_len24_prior_program_exact": 0.0,
      "standard_len24_oracle_program_exact": 0.0,
      "standard_len24_selected_state_prefix_fraction": 0.0026041666666666665,
      "standard_len24_oracle_state_prefix_fraction": 0.0013020833333333333,
      "standard_len24_avg_distinct_answers": 2.0,
      "standard_len24_selector_entropy": 1.385882019996643,
      "standard_len24_selected_oracle_gap_recovered": NaN,
      "paraphrase_len12_n": 32.0,
      "paraphrase_len12_beam_count": 4.0,
      "paraphrase_len12_beam0_accuracy": 0.03125,
      "paraphrase_len12_selected_accuracy": 0.03125,
      "paraphrase_len12_prior_accuracy": 0.0,
      "paraphrase_len12_oracle_accuracy": 0.03125,
      "paraphrase_len12_beam0_program_exact": 0.0,
      "paraphrase_len12_selected_program_exact": 0.0,
      "paraphrase_len12_prior_program_exact": 0.0,
      "paraphrase_len12_oracle_program_exact": 0.0,
      "paraphrase_len12_selected_state_prefix_fraction": 0.0,
      "paraphrase_len12_oracle_state_prefix_fraction": 0.0,
      "paraphrase_len12_avg_distinct_answers": 2.0,
      "paraphrase_len12_selector_entropy": 1.3859239295125008,
      "paraphrase_len12_selected_oracle_gap_recovered": NaN,
      "paraphrase_len24_n": 32.0,
      "paraphrase_len24_beam_count": 4.0,
      "paraphrase_len24_beam0_accuracy": 0.03125,
      "paraphrase_len24_selected_accuracy": 0.03125,
      "paraphrase_len24_prior_accuracy": 0.0625,
      "paraphrase_len24_oracle_accuracy": 0.09375,
      "paraphrase_len24_beam0_program_exact": 0.0,
      "paraphrase_len24_selected_program_exact": 0.0,
      "paraphrase_len24_prior_program_exact": 0.0,
      "paraphrase_len24_oracle_program_exact": 0.0,
      "paraphrase_len24_selected_state_prefix_fraction": 0.0,
      "paraphrase_len24_oracle_state_prefix_fraction": 0.0013020833333333333,
      "paraphrase_len24_avg_distinct_answers": 2.0,
      "paraphrase_len24_selector_entropy": 1.385916218161583,
      "paraphrase_len24_selected_oracle_gap_recovered": 0.0,
      "paired_len12_n": 64.0,
      "paired_len12_beam_count": 4.0,
      "paired_len12_beam0_accuracy": 0.0,
      "paired_len12_selected_accuracy": 0.0,
      "paired_len12_prior_accuracy": 0.0,
      "paired_len12_oracle_accuracy": 0.0,
      "paired_len12_beam0_program_exact": 0.0,
      "paired_len12_selected_program_exact": 0.0,
      "paired_len12_prior_program_exact": 0.0,
      "paired_len12_oracle_program_exact": 0.0,
      "paired_len12_selected_state_prefix_fraction": 0.0026041666666666665,
      "paired_len12_oracle_state_prefix_fraction": 0.0,
      "paired_len12_avg_distinct_answers": 2.0,
      "paired_len12_selector_entropy": 1.3859088085591793,
      "paired_len12_selected_oracle_gap_recovered": NaN,
      "paired_len12_selected_pair_answer_consistency": 1.0,
      "paired_len12_selected_pair_both_correct": 0.0,
      "paired_len12_oracle_pair_both_correct": 0.0,
      "paired_len12_selected_pair_state_consistency": 1.0,
      "paired_len12_oracle_pair_state_consistency": 1.0,
      "paired_len24_n": 64.0,
      "paired_len24_beam_count": 4.0,
      "paired_len24_beam0_accuracy": 0.03125,
      "paired_len24_selected_accuracy": 0.03125,
      "paired_len24_prior_accuracy": 0.0,
      "paired_len24_oracle_accuracy": 0.03125,
      "paired_len24_beam0_program_exact": 0.0,
      "paired_len24_selected_program_exact": 0.0,
      "paired_len24_prior_program_exact": 0.0,
      "paired_len24_oracle_program_exact": 0.0,
      "paired_len24_selected_state_prefix_fraction": 0.0,
      "paired_len24_oracle_state_prefix_fraction": 0.0,
      "paired_len24_avg_distinct_answers": 2.0,
      "paired_len24_selector_entropy": 1.3859056048095226,
      "paired_len24_selected_oracle_gap_recovered": NaN,
      "paired_len24_selected_pair_answer_consistency": 1.0,
      "paired_len24_selected_pair_both_correct": 0.03125,
      "paired_len24_oracle_pair_both_correct": 0.03125,
      "paired_len24_selected_pair_state_consistency": 1.0,
      "paired_len24_oracle_pair_state_consistency": 0.96875
    },
    {
      "stage": "medium",
      "local_step": 30,
      "step": 60,
      "loss": 42.217681884765625,
      "trace_set_loss": 33.493995666503906,
      "executor_loss": 4.5900068283081055,
      "state_set_loss": 3.8967700004577637,
      "selector_loss": 1.3560795783996582,
      "diversity_loss": 0.010312444530427456,
      "pair_set_loss": -0.692399799823761,
      "selector_matches_best_trace": 1.0,
      "best_soft_answer_mass": 0.010254206135869026,
      "mean_soft_answer_mass": 0.010152788832783699,
      "best_target_prob_beam_fraction": 0.0,
      "standard_len12_n": 32.0,
      "standard_len12_beam_count": 4.0,
      "standard_len12_beam0_accuracy": 0.03125,
      "standard_len12_selected_accuracy": 0.03125,
      "standard_len12_prior_accuracy": 0.03125,
      "standard_len12_oracle_accuracy": 0.0625,
      "standard_len12_beam0_program_exact": 0.0,
      "standard_len12_selected_program_exact": 0.0,
      "standard_len12_prior_program_exact": 0.0,
      "standard_len12_oracle_program_exact": 0.0,
      "standard_len12_selected_state_prefix_fraction": 0.0026041666666666665,
      "standard_len12_oracle_state_prefix_fraction": 0.0026041666666666665,
      "standard_len12_avg_distinct_answers": 2.0,
      "standard_len12_selector_entropy": 1.3856742903590202,
      "standard_len12_selected_oracle_gap_recovered": 0.0,
      "standard_len24_n": 32.0,
      "standard_len24_beam_count": 4.0,
      "standard_len24_beam0_accuracy": 0.0,
      "standard_len24_selected_accuracy": 0.0,
      "standard_len24_prior_accuracy": 0.03125,
      "standard_len24_oracle_accuracy": 0.03125,
      "standard_len24_beam0_program_exact": 0.0,
      "standard_len24_selected_program_exact": 0.0,
      "standard_len24_prior_program_exact": 0.0,
      "standard_len24_oracle_program_exact": 0.0,
      "standard_len24_selected_state_prefix_fraction": 0.0026041666666666665,
      "standard_len24_oracle_state_prefix_fraction": 0.0026041666666666665,
      "standard_len24_avg_distinct_answers": 2.0,
      "standard_len24_selector_entropy": 1.3855696693062782,
      "standard_len24_selected_oracle_gap_recovered": 0.0,
      "paraphrase_len12_n": 32.0,
      "paraphrase_len12_beam_count": 4.0,
      "paraphrase_len12_beam0_accuracy": 0.03125,
      "paraphrase_len12_selected_accuracy": 0.03125,
      "paraphrase_len12_prior_accuracy": 0.0,
      "paraphrase_len12_oracle_accuracy": 0.03125,
      "paraphrase_len12_beam0_program_exact": 0.0,
      "paraphrase_len12_selected_program_exact": 0.0,
      "paraphrase_len12_prior_program_exact": 0.0,
      "paraphrase_len12_oracle_program_exact": 0.0,
      "paraphrase_len12_selected_state_prefix_fraction": 0.0,
      "paraphrase_len12_oracle_state_prefix_fraction": 0.0,
      "paraphrase_len12_avg_distinct_answers": 2.0,
      "paraphrase_len12_selector_entropy": 1.3853870779275894,
      "paraphrase_len12_selected_oracle_gap_recovered": NaN,
      "paraphrase_len24_n": 32.0,
      "paraphrase_len24_beam_count": 4.0,
      "paraphrase_len24_beam0_accuracy": 0.0,
      "paraphrase_len24_selected_accuracy": 0.0,
      "paraphrase_len24_prior_accuracy": 0.0,
      "paraphrase_len24_oracle_accuracy": 0.0,
      "paraphrase_len24_beam0_program_exact": 0.0,
      "paraphrase_len24_selected_program_exact": 0.0,
      "paraphrase_len24_prior_program_exact": 0.0,
      "paraphrase_len24_oracle_program_exact": 0.0,
      "paraphrase_len24_selected_state_prefix_fraction": 0.0,
      "paraphrase_len24_oracle_state_prefix_fraction": 0.0,
      "paraphrase_len24_avg_distinct_answers": 2.0,
      "paraphrase_len24_selector_entropy": 1.385358527302742,
      "paraphrase_len24_selected_oracle_gap_recovered": NaN,
      "paired_len12_n": 64.0,
      "paired_len12_beam_count": 4.0,
      "paired_len12_beam0_accuracy": 0.015625,
      "paired_len12_selected_accuracy": 0.015625,
      "paired_len12_prior_accuracy": 0.0,
      "paired_len12_oracle_accuracy": 0.015625,
      "paired_len12_beam0_program_exact": 0.0,
      "paired_len12_selected_program_exact": 0.0,
      "paired_len12_prior_program_exact": 0.0,
      "paired_len12_oracle_program_exact": 0.0,
      "paired_len12_selected_state_prefix_fraction": 0.0,
      "paired_len12_oracle_state_prefix_fraction": 0.0,
      "paired_len12_avg_distinct_answers": 2.0,
      "paired_len12_selector_entropy": 1.3855420053005219,
      "paired_len12_selected_oracle_gap_recovered": NaN,
      "paired_len12_selected_pair_answer_consistency": 0.28125,
      "paired_len12_selected_pair_both_correct": 0.0,
      "paired_len12_oracle_pair_both_correct": 0.0,
      "paired_len12_selected_pair_state_consistency": 0.28125,
      "paired_len12_oracle_pair_state_consistency": 0.28125,
      "paired_len24_n": 64.0,
      "paired_len24_beam_count": 4.0,
      "paired_len24_beam0_accuracy": 0.046875,
      "paired_len24_selected_accuracy": 0.046875,
      "paired_len24_prior_accuracy": 0.0,
      "paired_len24_oracle_accuracy": 0.046875,
      "paired_len24_beam0_program_exact": 0.0,
      "paired_len24_selected_program_exact": 0.0,
      "paired_len24_prior_program_exact": 0.0,
      "paired_len24_oracle_program_exact": 0.0,
      "paired_len24_selected_state_prefix_fraction": 0.0,
      "paired_len24_oracle_state_prefix_fraction": 0.0,
      "paired_len24_avg_distinct_answers": 2.0,
      "paired_len24_selector_entropy": 1.3854734674096107,
      "paired_len24_selected_oracle_gap_recovered": NaN,
      "paired_len24_selected_pair_answer_consistency": 0.40625,
      "paired_len24_selected_pair_both_correct": 0.03125,
      "paired_len24_oracle_pair_both_correct": 0.03125,
      "paired_len24_selected_pair_state_consistency": 0.40625,
      "paired_len24_oracle_pair_state_consistency": 0.40625
    },
    {
      "stage": "train",
      "local_step": 1,
      "step": 61,
      "loss": 44.073768615722656,
      "trace_set_loss": 35.378265380859375,
      "executor_loss": 4.577037811279297,
      "state_set_loss": 3.885348320007324,
      "selector_loss": 1.3373252153396606,
      "diversity_loss": 0.010312624275684357,
      "pair_set_loss": -0.6931447982788086,
      "selector_matches_best_trace": 1.0,
      "best_soft_answer_mass": 0.010288206860423088,
      "mean_soft_answer_mass": 0.01028531789779663,
      "best_target_prob_beam_fraction": 1.0,
      "standard_len12_n": 32.0,
      "standard_len12_beam_count": 4.0,
      "standard_len12_beam0_accuracy": 0.0,
      "standard_len12_selected_accuracy": 0.0,
      "standard_len12_prior_accuracy": 0.03125,
      "standard_len12_oracle_accuracy": 0.03125,
      "standard_len12_beam0_program_exact": 0.0,
      "standard_len12_selected_program_exact": 0.0,
      "standard_len12_prior_program_exact": 0.0,
      "standard_len12_oracle_program_exact": 0.0,
      "standard_len12_selected_state_prefix_fraction": 0.0026041666666666665,
      "standard_len12_oracle_state_prefix_fraction": 0.0026041666666666665,
      "standard_len12_avg_distinct_answers": 2.0,
      "standard_len12_selector_entropy": 1.3854562565684319,
      "standard_len12_selected_oracle_gap_recovered": 0.0,
      "standard_len24_n": 32.0,
      "standard_len24_beam_count": 4.0,
      "standard_len24_beam0_accuracy": 0.0,
      "standard_len24_selected_accuracy": 0.0,
      "standard_len24_prior_accuracy": 0.03125,
      "standard_len24_oracle_accuracy": 0.03125,
      "standard_len24_beam0_program_exact": 0.0,
      "standard_len24_selected_program_exact": 0.0,
      "standard_len24_prior_program_exact": 0.0,
      "standard_len24_oracle_program_exact": 0.0,
      "standard_len24_selected_state_prefix_fraction": 0.0026041666666666665,
      "standard_len24_oracle_state_prefix_fraction": 0.0026041666666666665,
      "standard_len24_avg_distinct_answers": 2.0,
      "standard_len24_selector_entropy": 1.38536087423563,
      "standard_len24_selected_oracle_gap_recovered": 0.0,
      "paraphrase_len12_n": 32.0,
      "paraphrase_len12_beam_count": 4.0,
      "paraphrase_len12_beam0_accuracy": 0.03125,
      "paraphrase_len12_selected_accuracy": 0.03125,
      "paraphrase_len12_prior_accuracy": 0.0,
      "paraphrase_len12_oracle_accuracy": 0.03125,
      "paraphrase_len12_beam0_program_exact": 0.0,
      "paraphrase_len12_selected_program_exact": 0.0,
      "paraphrase_len12_prior_program_exact": 0.0,
      "paraphrase_len12_oracle_program_exact": 0.0,
      "paraphrase_len12_selected_state_prefix_fraction": 0.0,
      "paraphrase_len12_oracle_state_prefix_fraction": 0.0,
      "paraphrase_len12_avg_distinct_answers": 2.0,
      "paraphrase_len12_selector_entropy": 1.3851514011621475,
      "paraphrase_len12_selected_oracle_gap_recovered": NaN,
      "paraphrase_len24_n": 32.0,
      "paraphrase_len24_beam_count": 4.0,
      "paraphrase_len24_beam0_accuracy": 0.0,
      "paraphrase_len24_selected_accuracy": 0.0,
      "paraphrase_len24_prior_accuracy": 0.0,
      "paraphrase_len24_oracle_accuracy": 0.0,
      "paraphrase_len24_beam0_program_exact": 0.0,
      "paraphrase_len24_selected_program_exact": 0.0,
      "paraphrase_len24_prior_program_exact": 0.0,
      "paraphrase_len24_oracle_program_exact": 0.0,
      "paraphrase_len24_selected_state_prefix_fraction": 0.0,
      "paraphrase_len24_oracle_state_prefix_fraction": 0.0,
      "paraphrase_len24_avg_distinct_answers": 2.0,
      "paraphrase_len24_selector_entropy": 1.385127067565918,
      "paraphrase_len24_selected_oracle_gap_recovered": NaN,
      "paired_len12_n": 64.0,
      "paired_len12_beam_count": 4.0,
      "paired_len12_beam0_accuracy": 0.015625,
      "paired_len12_selected_accuracy": 0.015625,
      "paired_len12_prior_accuracy": 0.0,
      "paired_len12_oracle_accuracy": 0.015625,
      "paired_len12_beam0_program_exact": 0.0,
      "paired_len12_selected_program_exact": 0.0,
      "paired_len12_prior_program_exact": 0.0,
      "paired_len12_oracle_program_exact": 0.0,
      "paired_len12_selected_state_prefix_fraction": 0.0,
      "paired_len12_oracle_state_prefix_fraction": 0.0,
      "paired_len12_avg_distinct_answers": 2.0,
      "paired_len12_selector_entropy": 1.3853120878338814,
      "paired_len12_selected_oracle_gap_recovered": NaN,
      "paired_len12_selected_pair_answer_consistency": 0.3125,
      "paired_len12_selected_pair_both_correct": 0.0,
      "paired_len12_oracle_pair_both_correct": 0.0,
      "paired_len12_selected_pair_state_consistency": 0.3125,
      "paired_len12_oracle_pair_state_consistency": 0.28125,
      "paired_len24_n": 64.0,
      "paired_len24_beam_count": 4.0,
      "paired_len24_beam0_accuracy": 0.0625,
      "paired_len24_selected_accuracy": 0.0625,
      "paired_len24_prior_accuracy": 0.0,
      "paired_len24_oracle_accuracy": 0.0625,
      "paired_len24_beam0_program_exact": 0.0,
      "paired_len24_selected_program_exact": 0.0,
      "paired_len24_prior_program_exact": 0.0,
      "paired_len24_oracle_program_exact": 0.0,
      "paired_len24_selected_state_prefix_fraction": 0.0,
      "paired_len24_oracle_state_prefix_fraction": 0.0,
      "paired_len24_avg_distinct_answers": 2.0,
      "paired_len24_selector_entropy": 1.385256215929985,
      "paired_len24_selected_oracle_gap_recovered": NaN,
      "paired_len24_selected_pair_answer_consistency": 0.375,
      "paired_len24_selected_pair_both_correct": 0.0,
      "paired_len24_oracle_pair_both_correct": 0.0,
      "paired_len24_selected_pair_state_consistency": 0.375,
      "paired_len24_oracle_pair_state_consistency": 0.375
    },
    {
      "stage": "train",
      "local_step": 30,
      "step": 90,
      "loss": 42.98929214477539,
      "trace_set_loss": 34.28691101074219,
      "executor_loss": 4.575700759887695,
      "state_set_loss": 3.886500358581543,
      "selector_loss": 1.3726348876953125,
      "diversity_loss": 0.010309858247637749,
      "pair_set_loss": -0.6931471228599548,
      "selector_matches_best_trace": 0.5,
      "best_soft_answer_mass": 0.0102993194013834,
      "mean_soft_answer_mass": 0.01029907539486885,
      "best_target_prob_beam_fraction": 0.0,
      "standard_len12_n": 32.0,
      "standard_len12_beam_count": 4.0,
      "standard_len12_beam0_accuracy": 0.0,
      "standard_len12_selected_accuracy": 0.0,
      "standard_len12_prior_accuracy": 0.0,
      "standard_len12_oracle_accuracy": 0.0,
      "standard_len12_beam0_program_exact": 0.0,
      "standard_len12_selected_program_exact": 0.0,
      "standard_len12_prior_program_exact": 0.0,
      "standard_len12_oracle_program_exact": 0.0,
      "standard_len12_selected_state_prefix_fraction": 0.0,
      "standard_len12_oracle_state_prefix_fraction": 0.0,
      "standard_len12_avg_distinct_answers": 3.09375,
      "standard_len12_selector_entropy": 1.3811898529529572,
      "standard_len12_selected_oracle_gap_recovered": NaN,
      "standard_len24_n": 32.0,
      "standard_len24_beam_count": 4.0,
      "standard_len24_beam0_accuracy": 0.0,
      "standard_len24_selected_accuracy": 0.0,
      "standard_len24_prior_accuracy": 0.0,
      "standard_len24_oracle_accuracy": 0.0625,
      "standard_len24_beam0_program_exact": 0.0,
      "standard_len24_selected_program_exact": 0.0,
      "standard_len24_prior_program_exact": 0.0,
      "standard_len24_oracle_program_exact": 0.0,
      "standard_len24_selected_state_prefix_fraction": 0.0026041666666666665,
      "standard_len24_oracle_state_prefix_fraction": 0.0026041666666666665,
      "standard_len24_avg_distinct_answers": 3.09375,
      "standard_len24_selector_entropy": 1.3811540454626083,
      "standard_len24_selected_oracle_gap_recovered": 0.0,
      "paraphrase_len12_n": 32.0,
      "paraphrase_len12_beam_count": 4.0,
      "paraphrase_len12_beam0_accuracy": 0.0,
      "paraphrase_len12_selected_accuracy": 0.0,
      "paraphrase_len12_prior_accuracy": 0.0,
      "paraphrase_len12_oracle_accuracy": 0.0625,
      "paraphrase_len12_beam0_program_exact": 0.0,
      "paraphrase_len12_selected_program_exact": 0.0,
      "paraphrase_len12_prior_program_exact": 0.0,
      "paraphrase_len12_oracle_program_exact": 0.0,
      "paraphrase_len12_selected_state_prefix_fraction": 0.0026041666666666665,
      "paraphrase_len12_oracle_state_prefix_fraction": 0.0026041666666666665,
      "paraphrase_len12_avg_distinct_answers": 2.9375,
      "paraphrase_len12_selector_entropy": 1.3812540993094444,
      "paraphrase_len12_selected_oracle_gap_recovered": 0.0,
      "paraphrase_len24_n": 32.0,
      "paraphrase_len24_beam_count": 4.0,
      "paraphrase_len24_beam0_accuracy": 0.03125,
      "paraphrase_len24_selected_accuracy": 0.03125,
      "paraphrase_len24_prior_accuracy": 0.03125,
      "paraphrase_len24_oracle_accuracy": 0.03125,
      "paraphrase_len24_beam0_program_exact": 0.0,
      "paraphrase_len24_selected_program_exact": 0.0,
      "paraphrase_len24_prior_program_exact": 0.0,
      "paraphrase_len24_oracle_program_exact": 0.0,
      "paraphrase_len24_selected_state_prefix_fraction": 0.0013020833333333333,
      "paraphrase_len24_oracle_state_prefix_fraction": 0.0013020833333333333,
      "paraphrase_len24_avg_distinct_answers": 2.65625,
      "paraphrase_len24_selector_entropy": 1.3813889548182487,
      "paraphrase_len24_selected_oracle_gap_recovered": NaN,
      "paired_len12_n": 64.0,
      "paired_len12_beam_count": 4.0,
      "paired_len12_beam0_accuracy": 0.046875,
      "paired_len12_selected_accuracy": 0.046875,
      "paired_len12_prior_accuracy": 0.046875,
      "paired_len12_oracle_accuracy": 0.09375,
      "paired_len12_beam0_program_exact": 0.0,
      "paired_len12_selected_program_exact": 0.0,
      "paired_len12_prior_program_exact": 0.0,
      "paired_len12_oracle_program_exact": 0.0,
      "paired_len12_selected_state_prefix_fraction": 0.0013020833333333333,
      "paired_len12_oracle_state_prefix_fraction": 0.0,
      "paired_len12_avg_distinct_answers": 3.015625,
      "paired_len12_selector_entropy": 1.3811798207461834,
      "paired_len12_selected_oracle_gap_recovered": 0.0,
      "paired_len12_selected_pair_answer_consistency": 0.125,
      "paired_len12_selected_pair_both_correct": 0.03125,
      "paired_len12_oracle_pair_both_correct": 0.0625,
      "paired_len12_selected_pair_state_consistency": 0.0625,
      "paired_len12_oracle_pair_state_consistency": 0.15625,
      "paired_len24_n": 64.0,
      "paired_len24_beam_count": 4.0,
      "paired_len24_beam0_accuracy": 0.03125,
      "paired_len24_selected_accuracy": 0.03125,
      "paired_len24_prior_accuracy": 0.03125,
      "paired_len24_oracle_accuracy": 0.0625,
      "paired_len24_beam0_program_exact": 0.0,
      "paired_len24_selected_program_exact": 0.0,
      "paired_len24_prior_program_exact": 0.0,
      "paired_len24_oracle_program_exact": 0.0,
      "paired_len24_selected_state_prefix_fraction": 0.0,
      "paired_len24_oracle_state_prefix_fraction": 0.0,
      "paired_len24_avg_distinct_answers": 2.96875,
      "paired_len24_selector_entropy": 1.3813280947506428,
      "paired_len24_selected_oracle_gap_recovered": 0.0,
      "paired_len24_selected_pair_answer_consistency": 0.03125,
      "paired_len24_selected_pair_both_correct": 0.0,
      "paired_len24_oracle_pair_both_correct": 0.03125,
      "paired_len24_selected_pair_state_consistency": 0.0,
      "paired_len24_oracle_pair_state_consistency": 0.125
    },
    {
      "stage": "long",
      "local_step": 1,
      "step": 91,
      "loss": 40.789024353027344,
      "trace_set_loss": 32.07434844970703,
      "executor_loss": 4.575474739074707,
      "state_set_loss": 3.874269485473633,
      "selector_loss": 1.4963948726654053,
      "diversity_loss": 0.010309835895895958,
      "pair_set_loss": -0.6931471824645996,
      "selector_matches_best_trace": 0.0,
      "best_soft_answer_mass": 0.010301629081368446,
      "mean_soft_answer_mass": 0.010301409289240837,
      "best_target_prob_beam_fraction": 0.0,
      "standard_len12_n": 32.0,
      "standard_len12_beam_count": 4.0,
      "standard_len12_beam0_accuracy": 0.03125,
      "standard_len12_selected_accuracy": 0.03125,
      "standard_len12_prior_accuracy": 0.03125,
      "standard_len12_oracle_accuracy": 0.03125,
      "standard_len12_beam0_program_exact": 0.0,
      "standard_len12_selected_program_exact": 0.0,
      "standard_len12_prior_program_exact": 0.0,
      "standard_len12_oracle_program_exact": 0.0,
      "standard_len12_selected_state_prefix_fraction": 0.0,
      "standard_len12_oracle_state_prefix_fraction": 0.0,
      "standard_len12_avg_distinct_answers": 2.0,
      "standard_len12_selector_entropy": 1.380991391837597,
      "standard_len12_selected_oracle_gap_recovered": NaN,
      "standard_len24_n": 32.0,
      "standard_len24_beam_count": 4.0,
      "standard_len24_beam0_accuracy": 0.03125,
      "standard_len24_selected_accuracy": 0.03125,
      "standard_len24_prior_accuracy": 0.03125,
      "standard_len24_oracle_accuracy": 0.03125,
      "standard_len24_beam0_program_exact": 0.0,
      "standard_len24_selected_program_exact": 0.0,
      "standard_len24_prior_program_exact": 0.0,
      "standard_len24_oracle_program_exact": 0.0,
      "standard_len24_selected_state_prefix_fraction": 0.0013020833333333333,
      "standard_len24_oracle_state_prefix_fraction": 0.0013020833333333333,
      "standard_len24_avg_distinct_answers": 2.0,
      "standard_len24_selector_entropy": 1.380954921245575,
      "standard_len24_selected_oracle_gap_recovered": NaN,
      "paraphrase_len12_n": 32.0,
      "paraphrase_len12_beam_count": 4.0,
      "paraphrase_len12_beam0_accuracy": 0.0,
      "paraphrase_len12_selected_accuracy": 0.0,
      "paraphrase_len12_prior_accuracy": 0.0,
      "paraphrase_len12_oracle_accuracy": 0.03125,
      "paraphrase_len12_beam0_program_exact": 0.0,
      "paraphrase_len12_selected_program_exact": 0.0,
      "paraphrase_len12_prior_program_exact": 0.0,
      "paraphrase_len12_oracle_program_exact": 0.0,
      "paraphrase_len12_selected_state_prefix_fraction": 0.0026041666666666665,
      "paraphrase_len12_oracle_state_prefix_fraction": 0.005208333333333333,
      "paraphrase_len12_avg_distinct_answers": 2.0,
      "paraphrase_len12_selector_entropy": 1.3811045065522194,
      "paraphrase_len12_selected_oracle_gap_recovered": 0.0,
      "paraphrase_len24_n": 32.0,
      "paraphrase_len24_beam_count": 4.0,
      "paraphrase_len24_beam0_accuracy": 0.0,
      "paraphrase_len24_selected_accuracy": 0.0,
      "paraphrase_len24_prior_accuracy": 0.0,
      "paraphrase_len24_oracle_accuracy": 0.0,
      "paraphrase_len24_beam0_program_exact": 0.0,
      "paraphrase_len24_selected_program_exact": 0.0,
      "paraphrase_len24_prior_program_exact": 0.0,
      "paraphrase_len24_oracle_program_exact": 0.0,
      "paraphrase_len24_selected_state_prefix_fraction": 0.0013020833333333333,
      "paraphrase_len24_oracle_state_prefix_fraction": 0.0013020833333333333,
      "paraphrase_len24_avg_distinct_answers": 2.0,
      "paraphrase_len24_selector_entropy": 1.3812566250562668,
      "paraphrase_len24_selected_oracle_gap_recovered": NaN,
      "paired_len12_n": 64.0,
      "paired_len12_beam_count": 4.0,
      "paired_len12_beam0_accuracy": 0.015625,
      "paired_len12_selected_accuracy": 0.015625,
      "paired_len12_prior_accuracy": 0.015625,
      "paired_len12_oracle_accuracy": 0.015625,
      "paired_len12_beam0_program_exact": 0.0,
      "paired_len12_selected_program_exact": 0.0,
      "paired_len12_prior_program_exact": 0.0,
      "paired_len12_oracle_program_exact": 0.0,
      "paired_len12_selected_state_prefix_fraction": 0.0013020833333333333,
      "paired_len12_oracle_state_prefix_fraction": 0.0,
      "paired_len12_avg_distinct_answers": 2.0,
      "paired_len12_selector_entropy": 1.3810064941644669,
      "paired_len12_selected_oracle_gap_recovered": NaN,
      "paired_len12_selected_pair_answer_consistency": 0.15625,
      "paired_len12_selected_pair_both_correct": 0.0,
      "paired_len12_oracle_pair_both_correct": 0.0,
      "paired_len12_selected_pair_state_consistency": 0.125,
      "paired_len12_oracle_pair_state_consistency": 0.65625,
      "paired_len24_n": 64.0,
      "paired_len24_beam_count": 4.0,
      "paired_len24_beam0_accuracy": 0.015625,
      "paired_len24_selected_accuracy": 0.015625,
      "paired_len24_prior_accuracy": 0.015625,
      "paired_len24_oracle_accuracy": 0.046875,
      "paired_len24_beam0_program_exact": 0.0,
      "paired_len24_selected_program_exact": 0.0,
      "paired_len24_prior_program_exact": 0.0,
      "paired_len24_oracle_program_exact": 0.0,
      "paired_len24_selected_state_prefix_fraction": 0.0,
      "paired_len24_oracle_state_prefix_fraction": 0.0,
      "paired_len24_avg_distinct_answers": 2.0,
      "paired_len24_selector_entropy": 1.3811623826622963,
      "paired_len24_selected_oracle_gap_recovered": 0.0,
      "paired_len24_selected_pair_answer_consistency": 0.09375,
      "paired_len24_selected_pair_both_correct": 0.0,
      "paired_len24_oracle_pair_both_correct": 0.03125,
      "paired_len24_selected_pair_state_consistency": 0.0,
      "paired_len24_oracle_pair_state_consistency": 0.15625
    },
    {
      "stage": "long",
      "local_step": 30,
      "step": 120,
      "loss": 44.78605651855469,
      "trace_set_loss": 36.06793212890625,
      "executor_loss": 4.575623512268066,
      "state_set_loss": 3.8805012702941895,
      "selector_loss": 1.4817439317703247,
      "diversity_loss": 0.010309519246220589,
      "pair_set_loss": -0.6931471228599548,
      "selector_matches_best_trace": 0.0,
      "best_soft_answer_mass": 0.010300281457602978,
      "mean_soft_answer_mass": 0.010299877263605595,
      "best_target_prob_beam_fraction": 0.0,
      "standard_len12_n": 32.0,
      "standard_len12_beam_count": 4.0,
      "standard_len12_beam0_accuracy": 0.0,
      "standard_len12_selected_accuracy": 0.0,
      "standard_len12_prior_accuracy": 0.0,
      "standard_len12_oracle_accuracy": 0.03125,
      "standard_len12_beam0_program_exact": 0.0,
      "standard_len12_selected_program_exact": 0.0,
      "standard_len12_prior_program_exact": 0.0,
      "standard_len12_oracle_program_exact": 0.0,
      "standard_len12_selected_state_prefix_fraction": 0.0026041666666666665,
      "standard_len12_oracle_state_prefix_fraction": 0.0,
      "standard_len12_avg_distinct_answers": 2.0,
      "standard_len12_selector_entropy": 1.3735490590333939,
      "standard_len12_selected_oracle_gap_recovered": 0.0,
      "standard_len24_n": 32.0,
      "standard_len24_beam_count": 4.0,
      "standard_len24_beam0_accuracy": 0.0,
      "standard_len24_selected_accuracy": 0.0,
      "standard_len24_prior_accuracy": 0.0,
      "standard_len24_oracle_accuracy": 0.0,
      "standard_len24_beam0_program_exact": 0.0,
      "standard_len24_selected_program_exact": 0.0,
      "standard_len24_prior_program_exact": 0.0,
      "standard_len24_oracle_program_exact": 0.0,
      "standard_len24_selected_state_prefix_fraction": 0.0013020833333333333,
      "standard_len24_oracle_state_prefix_fraction": 0.0013020833333333333,
      "standard_len24_avg_distinct_answers": 2.34375,
      "standard_len24_selector_entropy": 1.3731747046113014,
      "standard_len24_selected_oracle_gap_recovered": NaN,
      "paraphrase_len12_n": 32.0,
      "paraphrase_len12_beam_count": 4.0,
      "paraphrase_len12_beam0_accuracy": 0.03125,
      "paraphrase_len12_selected_accuracy": 0.03125,
      "paraphrase_len12_prior_accuracy": 0.03125,
      "paraphrase_len12_oracle_accuracy": 0.09375,
      "paraphrase_len12_beam0_program_exact": 0.0,
      "paraphrase_len12_selected_program_exact": 0.0,
      "paraphrase_len12_prior_program_exact": 0.0,
      "paraphrase_len12_oracle_program_exact": 0.0,
      "paraphrase_len12_selected_state_prefix_fraction": 0.0,
      "paraphrase_len12_oracle_state_prefix_fraction": 0.0,
      "paraphrase_len12_avg_distinct_answers": 2.71875,
      "paraphrase_len12_selector_entropy": 1.3732820451259613,
      "paraphrase_len12_selected_oracle_gap_recovered": 0.0,
      "paraphrase_len24_n": 32.0,
      "paraphrase_len24_beam_count": 4.0,
      "paraphrase_len24_beam0_accuracy": 0.0,
      "paraphrase_len24_selected_accuracy": 0.0,
      "paraphrase_len24_prior_accuracy": 0.0,
      "paraphrase_len24_oracle_accuracy": 0.0,
      "paraphrase_len24_beam0_program_exact": 0.0,
      "paraphrase_len24_selected_program_exact": 0.0,
      "paraphrase_len24_prior_program_exact": 0.0,
      "paraphrase_len24_oracle_program_exact": 0.0,
      "paraphrase_len24_selected_state_prefix_fraction": 0.0013020833333333333,
      "paraphrase_len24_oracle_state_prefix_fraction": 0.0013020833333333333,
      "paraphrase_len24_avg_distinct_answers": 3.75,
      "paraphrase_len24_selector_entropy": 1.3728632554411888,
      "paraphrase_len24_selected_oracle_gap_recovered": NaN,
      "paired_len12_n": 64.0,
      "paired_len12_beam_count": 4.0,
      "paired_len12_beam0_accuracy": 0.015625,
      "paired_len12_selected_accuracy": 0.015625,
      "paired_len12_prior_accuracy": 0.015625,
      "paired_len12_oracle_accuracy": 0.015625,
      "paired_len12_beam0_program_exact": 0.0,
      "paired_len12_selected_program_exact": 0.0,
      "paired_len12_prior_program_exact": 0.0,
      "paired_len12_oracle_program_exact": 0.0,
      "paired_len12_selected_state_prefix_fraction": 0.0026041666666666665,
      "paired_len12_oracle_state_prefix_fraction": 0.0,
      "paired_len12_avg_distinct_answers": 2.390625,
      "paired_len12_selector_entropy": 1.3732594661414623,
      "paired_len12_selected_oracle_gap_recovered": NaN,
      "paired_len12_selected_pair_answer_consistency": 0.59375,
      "paired_len12_selected_pair_both_correct": 0.0,
      "paired_len12_oracle_pair_both_correct": 0.0,
      "paired_len12_selected_pair_state_consistency": 0.59375,
      "paired_len12_oracle_pair_state_consistency": 0.21875,
      "paired_len24_n": 64.0,
      "paired_len24_beam_count": 4.0,
      "paired_len24_beam0_accuracy": 0.03125,
      "paired_len24_selected_accuracy": 0.03125,
      "paired_len24_prior_accuracy": 0.03125,
      "paired_len24_oracle_accuracy": 0.09375,
      "paired_len24_beam0_program_exact": 0.0,
      "paired_len24_selected_program_exact": 0.0,
      "paired_len24_prior_program_exact": 0.0,
      "paired_len24_oracle_program_exact": 0.0,
      "paired_len24_selected_state_prefix_fraction": 0.0,
      "paired_len24_oracle_state_prefix_fraction": 0.0,
      "paired_len24_avg_distinct_answers": 3.078125,
      "paired_len24_selector_entropy": 1.372943241149187,
      "paired_len24_selected_oracle_gap_recovered": 0.0,
      "paired_len24_selected_pair_answer_consistency": 0.15625,
      "paired_len24_selected_pair_both_correct": 0.0,
      "paired_len24_oracle_pair_both_correct": 0.03125,
      "paired_len24_selected_pair_state_consistency": 0.09375,
      "paired_len24_oracle_pair_state_consistency": 0.25
    }
  ],
  "final_metrics": {
    "standard_len12": {
      "n": 32.0,
      "beam_count": 4.0,
      "beam0_accuracy": 0.0,
      "selected_accuracy": 0.0,
      "prior_accuracy": 0.0,
      "oracle_accuracy": 0.03125,
      "beam0_program_exact": 0.0,
      "selected_program_exact": 0.0,
      "prior_program_exact": 0.0,
      "oracle_program_exact": 0.0,
      "selected_state_prefix_fraction": 0.0026041666666666665,
      "oracle_state_prefix_fraction": 0.0,
      "avg_distinct_answers": 2.0,
      "selector_entropy": 1.3735490590333939,
      "selected_oracle_gap_recovered": 0.0
    },
    "standard_len24": {
      "n": 32.0,
      "beam_count": 4.0,
      "beam0_accuracy": 0.0,
      "selected_accuracy": 0.0,
      "prior_accuracy": 0.0,
      "oracle_accuracy": 0.0,
      "beam0_program_exact": 0.0,
      "selected_program_exact": 0.0,
      "prior_program_exact": 0.0,
      "oracle_program_exact": 0.0,
      "selected_state_prefix_fraction": 0.0013020833333333333,
      "oracle_state_prefix_fraction": 0.0013020833333333333,
      "avg_distinct_answers": 2.34375,
      "selector_entropy": 1.3731747046113014,
      "selected_oracle_gap_recovered": NaN
    },
    "paraphrase_len12": {
      "n": 32.0,
      "beam_count": 4.0,
      "beam0_accuracy": 0.03125,
      "selected_accuracy": 0.03125,
      "prior_accuracy": 0.03125,
      "oracle_accuracy": 0.09375,
      "beam0_program_exact": 0.0,
      "selected_program_exact": 0.0,
      "prior_program_exact": 0.0,
      "oracle_program_exact": 0.0,
      "selected_state_prefix_fraction": 0.0,
      "oracle_state_prefix_fraction": 0.0,
      "avg_distinct_answers": 2.71875,
      "selector_entropy": 1.3732820451259613,
      "selected_oracle_gap_recovered": 0.0
    },
    "paraphrase_len24": {
      "n": 32.0,
      "beam_count": 4.0,
      "beam0_accuracy": 0.0,
      "selected_accuracy": 0.0,
      "prior_accuracy": 0.0,
      "oracle_accuracy": 0.0,
      "beam0_program_exact": 0.0,
      "selected_program_exact": 0.0,
      "prior_program_exact": 0.0,
      "oracle_program_exact": 0.0,
      "selected_state_prefix_fraction": 0.0013020833333333333,
      "oracle_state_prefix_fraction": 0.0013020833333333333,
      "avg_distinct_answers": 3.75,
      "selector_entropy": 1.3728632554411888,
      "selected_oracle_gap_recovered": NaN
    },
    "paired_len12": {
      "n": 64.0,
      "beam_count": 4.0,
      "beam0_accuracy": 0.015625,
      "selected_accuracy": 0.015625,
      "prior_accuracy": 0.015625,
      "oracle_accuracy": 0.015625,
      "beam0_program_exact": 0.0,
      "selected_program_exact": 0.0,
      "prior_program_exact": 0.0,
      "oracle_program_exact": 0.0,
      "selected_state_prefix_fraction": 0.0026041666666666665,
      "oracle_state_prefix_fraction": 0.0,
      "avg_distinct_answers": 2.390625,
      "selector_entropy": 1.3732594661414623,
      "selected_oracle_gap_recovered": NaN,
      "selected_pair_answer_consistency": 0.59375,
      "selected_pair_both_correct": 0.0,
      "oracle_pair_both_correct": 0.0,
      "selected_pair_state_consistency": 0.59375,
      "oracle_pair_state_consistency": 0.21875
    },
    "paired_len24": {
      "n": 64.0,
      "beam_count": 4.0,
      "beam0_accuracy": 0.03125,
      "selected_accuracy": 0.03125,
      "prior_accuracy": 0.03125,
      "oracle_accuracy": 0.09375,
      "beam0_program_exact": 0.0,
      "selected_program_exact": 0.0,
      "prior_program_exact": 0.0,
      "oracle_program_exact": 0.0,
      "selected_state_prefix_fraction": 0.0,
      "oracle_state_prefix_fraction": 0.0,
      "avg_distinct_answers": 3.078125,
      "selector_entropy": 1.372943241149187,
      "selected_oracle_gap_recovered": 0.0,
      "selected_pair_answer_consistency": 0.15625,
      "selected_pair_both_correct": 0.0,
      "oracle_pair_both_correct": 0.03125,
      "selected_pair_state_consistency": 0.09375,
      "oracle_pair_state_consistency": 0.25
    }
  }
}