{
  "args": {
    "model_id": "Qwen/Qwen3-4B",
    "load_in_4bit": true,
    "torch_dtype": "bf16",
    "device_map": "auto",
    "use_lora": true,
    "lora_r": 8,
    "lora_alpha": 16,
    "lora_dropout": 0.05,
    "lora_target_modules": "all-linear",
    "gradient_checkpointing": true,
    "optimizer": "paged_adamw_8bit",
    "modulus": 97,
    "max_steps": 24,
    "beam_count": 4,
    "beam_register_mode": "latent",
    "train_min_len": 1,
    "train_max_len": 12,
    "curriculum_stages": "short:1:4:40,medium:1:8:40,train:1:12:40,long:8:24:40",
    "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": 128,
    "eval_size": 64,
    "train_batch_size": 4,
    "eval_batch_size": 4,
    "eval_every": 80,
    "stage_eval_every": 40,
    "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": 1536,
    "seed": 13,
    "output_dir": "experiments/qwen_latent_beam_program_compiler/runs/pilot_latent_beam4_s160",
    "checkpoint_dir": "large_artifacts/qwen_latent_beam_program_compiler/checkpoints/pilot_latent_beam4_s160"
  },
  "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": 8,
    "gpu_name": "NVIDIA RTX 6000 Ada Generation",
    "gpu_vram_gb": 47.383
  },
  "hidden_dim": 2560,
  "beam_count": 4,
  "prompt_bank_count": 1,
  "beam_register_mode": "latent",
  "register_count": 49,
  "dataset": {
    "train_size": 128,
    "curriculum_stages": [
      {
        "name": "short",
        "min_len": 1,
        "max_len": 4,
        "steps": 40
      },
      {
        "name": "medium",
        "min_len": 1,
        "max_len": 8,
        "steps": 40
      },
      {
        "name": "train",
        "min_len": 1,
        "max_len": 12,
        "steps": 40
      },
      {
        "name": "long",
        "min_len": 8,
        "max_len": 24,
        "steps": 40
      }
    ],
    "eval_size": 64,
    "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": 838.4032125473022,
  "checkpoints": [
    "large_artifacts/qwen_latent_beam_program_compiler/checkpoints/pilot_latent_beam4_s160/adapter",
    "large_artifacts/qwen_latent_beam_program_compiler/checkpoints/pilot_latent_beam4_s160/heads.pt"
  ],
  "train_log": [
    {
      "stage": "short",
      "local_step": 1,
      "step": 1,
      "loss": 46.06187057495117,
      "trace_set_loss": 37.35139846801758,
      "executor_loss": 4.580686569213867,
      "state_set_loss": 3.8863754272460938,
      "selector_loss": 1.3887910842895508,
      "diversity_loss": 0.010351646691560745,
      "pair_set_loss": -0.6931462287902832,
      "selector_matches_best_trace": 0.25,
      "best_soft_answer_mass": 0.010249088518321514,
      "mean_soft_answer_mass": 0.010247859172523022,
      "best_target_prob_beam_fraction": 0.25,
      "standard_len12_n": 64.0,
      "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": 0.0026041666666666665,
      "standard_len12_avg_distinct_answers": 1.21875,
      "standard_len12_selector_entropy": 1.3862504065036774,
      "standard_len12_selected_oracle_gap_recovered": NaN,
      "standard_len24_n": 64.0,
      "standard_len24_beam_count": 4.0,
      "standard_len24_beam0_accuracy": 0.03125,
      "standard_len24_selected_accuracy": 0.03125,
      "standard_len24_prior_accuracy": 0.015625,
      "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": 2.8125,
      "standard_len24_selector_entropy": 1.386251650750637,
      "standard_len24_selected_oracle_gap_recovered": NaN,
      "paraphrase_len12_n": 64.0,
      "paraphrase_len12_beam_count": 4.0,
      "paraphrase_len12_beam0_accuracy": 0.015625,
      "paraphrase_len12_selected_accuracy": 0.015625,
      "paraphrase_len12_prior_accuracy": 0.03125,
      "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": 1.953125,
      "paraphrase_len12_selector_entropy": 1.3862487822771072,
      "paraphrase_len12_selected_oracle_gap_recovered": 0.0,
      "paraphrase_len24_n": 64.0,
      "paraphrase_len24_beam_count": 4.0,
      "paraphrase_len24_beam0_accuracy": 0.015625,
      "paraphrase_len24_selected_accuracy": 0.015625,
      "paraphrase_len24_prior_accuracy": 0.046875,
      "paraphrase_len24_oracle_accuracy": 0.0625,
      "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": 0.0006510416666666666,
      "paraphrase_len24_oracle_state_prefix_fraction": 0.0006510416666666666,
      "paraphrase_len24_avg_distinct_answers": 2.75,
      "paraphrase_len24_selector_entropy": 1.3862492814660072,
      "paraphrase_len24_selected_oracle_gap_recovered": 0.0,
      "paired_len12_n": 128.0,
      "paired_len12_beam_count": 4.0,
      "paired_len12_beam0_accuracy": 0.0234375,
      "paired_len12_selected_accuracy": 0.0234375,
      "paired_len12_prior_accuracy": 0.015625,
      "paired_len12_oracle_accuracy": 0.0234375,
      "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": 0.0013020833333333333,
      "paired_len12_avg_distinct_answers": 1.578125,
      "paired_len12_selector_entropy": 1.3862495720386505,
      "paired_len12_selected_oracle_gap_recovered": NaN,
      "paired_len12_selected_pair_answer_consistency": 0.0625,
      "paired_len12_selected_pair_both_correct": 0.015625,
      "paired_len12_oracle_pair_both_correct": 0.015625,
      "paired_len12_selected_pair_state_consistency": 0.03125,
      "paired_len12_oracle_pair_state_consistency": 0.015625,
      "paired_len24_n": 128.0,
      "paired_len24_beam_count": 4.0,
      "paired_len24_beam0_accuracy": 0.015625,
      "paired_len24_selected_accuracy": 0.015625,
      "paired_len24_prior_accuracy": 0.0,
      "paired_len24_oracle_accuracy": 0.0390625,
      "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": 0.0006510416666666666,
      "paired_len24_oracle_state_prefix_fraction": 0.0006510416666666666,
      "paired_len24_avg_distinct_answers": 2.8046875,
      "paired_len24_selector_entropy": 1.3862504735589027,
      "paired_len24_selected_oracle_gap_recovered": 0.0,
      "paired_len24_selected_pair_answer_consistency": 0.0,
      "paired_len24_selected_pair_both_correct": 0.0,
      "paired_len24_oracle_pair_both_correct": 0.0,
      "paired_len24_selected_pair_state_consistency": 0.0,
      "paired_len24_oracle_pair_state_consistency": 0.0
    },
    {
      "stage": "short",
      "local_step": 40,
      "step": 40,
      "loss": 38.69422912597656,
      "trace_set_loss": 30.000064849853516,
      "executor_loss": 4.57868766784668,
      "state_set_loss": 3.872394561767578,
      "selector_loss": 1.3871512413024902,
      "diversity_loss": 0.010322274640202522,
      "pair_set_loss": -0.6931469440460205,
      "selector_matches_best_trace": 0.25,
      "best_soft_answer_mass": 0.01026904210448265,
      "mean_soft_answer_mass": 0.01026836596429348,
      "best_target_prob_beam_fraction": 0.25,
      "standard_len12_n": 64.0,
      "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": 1.0,
      "standard_len12_selector_entropy": 1.3862817287445068,
      "standard_len12_selected_oracle_gap_recovered": NaN,
      "standard_len24_n": 64.0,
      "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": 1.0,
      "standard_len24_selector_entropy": 1.3862818107008934,
      "standard_len24_selected_oracle_gap_recovered": NaN,
      "paraphrase_len12_n": 64.0,
      "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": 0.0,
      "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": 1.0,
      "paraphrase_len12_selector_entropy": 1.3862816914916039,
      "paraphrase_len12_selected_oracle_gap_recovered": NaN,
      "paraphrase_len24_n": 64.0,
      "paraphrase_len24_beam_count": 4.0,
      "paraphrase_len24_beam0_accuracy": 0.015625,
      "paraphrase_len24_selected_accuracy": 0.015625,
      "paraphrase_len24_prior_accuracy": 0.015625,
      "paraphrase_len24_oracle_accuracy": 0.015625,
      "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": 1.0,
      "paraphrase_len24_selector_entropy": 1.3862817585468292,
      "paraphrase_len24_selected_oracle_gap_recovered": NaN,
      "paired_len12_n": 128.0,
      "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": 0.0013020833333333333,
      "paired_len12_avg_distinct_answers": 1.0,
      "paired_len12_selector_entropy": 1.3862817287445068,
      "paired_len12_selected_oracle_gap_recovered": NaN,
      "paired_len12_selected_pair_answer_consistency": 1.0,
      "paired_len12_selected_pair_both_correct": 0.015625,
      "paired_len12_oracle_pair_both_correct": 0.015625,
      "paired_len12_selected_pair_state_consistency": 1.0,
      "paired_len12_oracle_pair_state_consistency": 1.0,
      "paired_len24_n": 128.0,
      "paired_len24_beam_count": 4.0,
      "paired_len24_beam0_accuracy": 0.0,
      "paired_len24_selected_accuracy": 0.0,
      "paired_len24_prior_accuracy": 0.0,
      "paired_len24_oracle_accuracy": 0.0,
      "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": 1.0,
      "paired_len24_selector_entropy": 1.386281792074442,
      "paired_len24_selected_oracle_gap_recovered": NaN,
      "paired_len24_selected_pair_answer_consistency": 1.0,
      "paired_len24_selected_pair_both_correct": 0.0,
      "paired_len24_oracle_pair_both_correct": 0.0,
      "paired_len24_selected_pair_state_consistency": 1.0,
      "paired_len24_oracle_pair_state_consistency": 1.0
    },
    {
      "stage": "medium",
      "local_step": 1,
      "step": 41,
      "loss": 45.73719787597656,
      "trace_set_loss": 37.04991149902344,
      "executor_loss": 4.5784149169921875,
      "state_set_loss": 3.866271734237671,
      "selector_loss": 1.384743571281433,
      "diversity_loss": 0.010321259498596191,
      "pair_set_loss": -0.6931468844413757,
      "selector_matches_best_trace": 0.25,
      "best_soft_answer_mass": 0.010271711274981499,
      "mean_soft_answer_mass": 0.010271161794662476,
      "best_target_prob_beam_fraction": 0.25,
      "standard_len12_n": 64.0,
      "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": 0.00390625,
      "standard_len12_oracle_state_prefix_fraction": 0.00390625,
      "standard_len12_avg_distinct_answers": 1.0,
      "standard_len12_selector_entropy": 1.3862807676196098,
      "standard_len12_selected_oracle_gap_recovered": NaN,
      "standard_len24_n": 64.0,
      "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": 1.0,
      "standard_len24_selector_entropy": 1.386280968785286,
      "standard_len24_selected_oracle_gap_recovered": NaN,
      "paraphrase_len12_n": 64.0,
      "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": 0.0,
      "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": 1.0,
      "paraphrase_len12_selector_entropy": 1.386280745267868,
      "paraphrase_len12_selected_oracle_gap_recovered": NaN,
      "paraphrase_len24_n": 64.0,
      "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": 0.0006510416666666666,
      "paraphrase_len24_oracle_state_prefix_fraction": 0.0006510416666666666,
      "paraphrase_len24_avg_distinct_answers": 1.0,
      "paraphrase_len24_selector_entropy": 1.3862809091806412,
      "paraphrase_len24_selected_oracle_gap_recovered": NaN,
      "paired_len12_n": 128.0,
      "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": 0.0026041666666666665,
      "paired_len12_avg_distinct_answers": 1.0,
      "paired_len12_selector_entropy": 1.3862807638943195,
      "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": 128.0,
      "paired_len24_beam_count": 4.0,
      "paired_len24_beam0_accuracy": 0.0,
      "paired_len24_selected_accuracy": 0.0,
      "paired_len24_prior_accuracy": 0.0,
      "paired_len24_oracle_accuracy": 0.0,
      "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": 1.0,
      "paired_len24_selector_entropy": 1.3862809389829636,
      "paired_len24_selected_oracle_gap_recovered": NaN,
      "paired_len24_selected_pair_answer_consistency": 1.0,
      "paired_len24_selected_pair_both_correct": 0.0,
      "paired_len24_oracle_pair_both_correct": 0.0,
      "paired_len24_selected_pair_state_consistency": 1.0,
      "paired_len24_oracle_pair_state_consistency": 1.0
    },
    {
      "stage": "medium",
      "local_step": 40,
      "step": 80,
      "loss": 42.15321731567383,
      "trace_set_loss": 33.40963363647461,
      "executor_loss": 4.592494964599609,
      "state_set_loss": 3.9075942039489746,
      "selector_loss": 1.3892110586166382,
      "diversity_loss": 0.010327648371458054,
      "pair_set_loss": -0.6931445598602295,
      "selector_matches_best_trace": 0.0,
      "best_soft_answer_mass": 0.010134824551641941,
      "mean_soft_answer_mass": 0.010129081085324287,
      "best_target_prob_beam_fraction": 0.0,
      "standard_len12_n": 64.0,
      "standard_len12_beam_count": 4.0,
      "standard_len12_beam0_accuracy": 0.0,
      "standard_len12_selected_accuracy": 0.015625,
      "standard_len12_prior_accuracy": 0.015625,
      "standard_len12_oracle_accuracy": 0.015625,
      "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": 1.828125,
      "standard_len12_selector_entropy": 1.386289358139038,
      "standard_len12_selected_oracle_gap_recovered": 1.0,
      "standard_len24_n": 64.0,
      "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": 1.390625,
      "standard_len24_selector_entropy": 1.3862893655896187,
      "standard_len24_selected_oracle_gap_recovered": NaN,
      "paraphrase_len12_n": 64.0,
      "paraphrase_len12_beam_count": 4.0,
      "paraphrase_len12_beam0_accuracy": 0.015625,
      "paraphrase_len12_selected_accuracy": 0.0,
      "paraphrase_len12_prior_accuracy": 0.0,
      "paraphrase_len12_oracle_accuracy": 0.015625,
      "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": 1.28125,
      "paraphrase_len12_selector_entropy": 1.3862893655896187,
      "paraphrase_len12_selected_oracle_gap_recovered": NaN,
      "paraphrase_len24_n": 64.0,
      "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": 0.0006510416666666666,
      "paraphrase_len24_oracle_state_prefix_fraction": 0.0006510416666666666,
      "paraphrase_len24_avg_distinct_answers": 1.21875,
      "paraphrase_len24_selector_entropy": 1.3862893730401993,
      "paraphrase_len24_selected_oracle_gap_recovered": NaN,
      "paired_len12_n": 128.0,
      "paired_len12_beam_count": 4.0,
      "paired_len12_beam0_accuracy": 0.0546875,
      "paired_len12_selected_accuracy": 0.046875,
      "paired_len12_prior_accuracy": 0.046875,
      "paired_len12_oracle_accuracy": 0.0703125,
      "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": 0.0013020833333333333,
      "paired_len12_avg_distinct_answers": 1.5,
      "paired_len12_selector_entropy": 1.386289358139038,
      "paired_len12_selected_oracle_gap_recovered": -0.5,
      "paired_len12_selected_pair_answer_consistency": 0.65625,
      "paired_len12_selected_pair_both_correct": 0.03125,
      "paired_len12_oracle_pair_both_correct": 0.0625,
      "paired_len12_selected_pair_state_consistency": 0.65625,
      "paired_len12_oracle_pair_state_consistency": 0.421875,
      "paired_len24_n": 128.0,
      "paired_len24_beam_count": 4.0,
      "paired_len24_beam0_accuracy": 0.0234375,
      "paired_len24_selected_accuracy": 0.0234375,
      "paired_len24_prior_accuracy": 0.0234375,
      "paired_len24_oracle_accuracy": 0.0234375,
      "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": 1.2734375,
      "paired_len24_selector_entropy": 1.3862893655896187,
      "paired_len24_selected_oracle_gap_recovered": NaN,
      "paired_len24_selected_pair_answer_consistency": 0.3125,
      "paired_len24_selected_pair_both_correct": 0.0,
      "paired_len24_oracle_pair_both_correct": 0.0,
      "paired_len24_selected_pair_state_consistency": 0.3125,
      "paired_len24_oracle_pair_state_consistency": 0.203125
    },
    {
      "stage": "train",
      "local_step": 1,
      "step": 81,
      "loss": 46.95644760131836,
      "trace_set_loss": 38.250762939453125,
      "executor_loss": 4.577873229980469,
      "state_set_loss": 3.8848304748535156,
      "selector_loss": 1.386638879776001,
      "diversity_loss": 0.0103123988956213,
      "pair_set_loss": -0.6931470632553101,
      "selector_matches_best_trace": 0.25,
      "best_soft_answer_mass": 0.01027725450694561,
      "mean_soft_answer_mass": 0.01027673203498125,
      "best_target_prob_beam_fraction": 0.25,
      "standard_len12_n": 64.0,
      "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": 1.59375,
      "standard_len12_selector_entropy": 1.3862898349761963,
      "standard_len12_selected_oracle_gap_recovered": NaN,
      "standard_len24_n": 64.0,
      "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": 2.671875,
      "standard_len24_selector_entropy": 1.3862898349761963,
      "standard_len24_selected_oracle_gap_recovered": NaN,
      "paraphrase_len12_n": 64.0,
      "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": 0.0,
      "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": 1.671875,
      "paraphrase_len12_selector_entropy": 1.3862898349761963,
      "paraphrase_len12_selected_oracle_gap_recovered": NaN,
      "paraphrase_len24_n": 64.0,
      "paraphrase_len24_beam_count": 4.0,
      "paraphrase_len24_beam0_accuracy": 0.015625,
      "paraphrase_len24_selected_accuracy": 0.015625,
      "paraphrase_len24_prior_accuracy": 0.015625,
      "paraphrase_len24_oracle_accuracy": 0.0625,
      "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": 0.0006510416666666666,
      "paraphrase_len24_oracle_state_prefix_fraction": 0.0006510416666666666,
      "paraphrase_len24_avg_distinct_answers": 2.90625,
      "paraphrase_len24_selector_entropy": 1.3862898349761963,
      "paraphrase_len24_selected_oracle_gap_recovered": 0.0,
      "paired_len12_n": 128.0,
      "paired_len12_beam_count": 4.0,
      "paired_len12_beam0_accuracy": 0.046875,
      "paired_len12_selected_accuracy": 0.03125,
      "paired_len12_prior_accuracy": 0.03125,
      "paired_len12_oracle_accuracy": 0.0625,
      "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": 1.578125,
      "paired_len12_selector_entropy": 1.3862898349761963,
      "paired_len12_selected_oracle_gap_recovered": -1.0,
      "paired_len12_selected_pair_answer_consistency": 0.34375,
      "paired_len12_selected_pair_both_correct": 0.015625,
      "paired_len12_oracle_pair_both_correct": 0.046875,
      "paired_len12_selected_pair_state_consistency": 0.296875,
      "paired_len12_oracle_pair_state_consistency": 0.390625,
      "paired_len24_n": 128.0,
      "paired_len24_beam_count": 4.0,
      "paired_len24_beam0_accuracy": 0.015625,
      "paired_len24_selected_accuracy": 0.0,
      "paired_len24_prior_accuracy": 0.0,
      "paired_len24_oracle_accuracy": 0.015625,
      "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": 2.671875,
      "paired_len24_selector_entropy": 1.3862898349761963,
      "paired_len24_selected_oracle_gap_recovered": NaN,
      "paired_len24_selected_pair_answer_consistency": 0.09375,
      "paired_len24_selected_pair_both_correct": 0.0,
      "paired_len24_oracle_pair_both_correct": 0.0,
      "paired_len24_selected_pair_state_consistency": 0.0,
      "paired_len24_oracle_pair_state_consistency": 0.015625
    },
    {
      "stage": "train",
      "local_step": 40,
      "step": 120,
      "loss": 43.160667419433594,
      "trace_set_loss": 34.46480178833008,
      "executor_loss": 4.575519561767578,
      "state_set_loss": 3.8774285316467285,
      "selector_loss": 1.3863229751586914,
      "diversity_loss": 0.010310739278793335,
      "pair_set_loss": -0.6931471824645996,
      "selector_matches_best_trace": 0.0,
      "best_soft_answer_mass": 0.010301120579242706,
      "mean_soft_answer_mass": 0.010300945490598679,
      "best_target_prob_beam_fraction": 0.25,
      "standard_len12_n": 64.0,
      "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": 1.0,
      "standard_len12_selector_entropy": 1.3862933665513992,
      "standard_len12_selected_oracle_gap_recovered": NaN,
      "standard_len24_n": 64.0,
      "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": 1.0,
      "standard_len24_selector_entropy": 1.3862934112548828,
      "standard_len24_selected_oracle_gap_recovered": NaN,
      "paraphrase_len12_n": 64.0,
      "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": 0.03125,
      "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": 0.0013020833333333333,
      "paraphrase_len12_oracle_state_prefix_fraction": 0.0013020833333333333,
      "paraphrase_len12_avg_distinct_answers": 1.0,
      "paraphrase_len12_selector_entropy": 1.3862934038043022,
      "paraphrase_len12_selected_oracle_gap_recovered": NaN,
      "paraphrase_len24_n": 64.0,
      "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": 0.0006510416666666666,
      "paraphrase_len24_oracle_state_prefix_fraction": 0.0006510416666666666,
      "paraphrase_len24_avg_distinct_answers": 1.0,
      "paraphrase_len24_selector_entropy": 1.3862933441996574,
      "paraphrase_len24_selected_oracle_gap_recovered": NaN,
      "paired_len12_n": 128.0,
      "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": 0.0,
      "paired_len12_oracle_state_prefix_fraction": 0.0,
      "paired_len12_avg_distinct_answers": 1.0,
      "paired_len12_selector_entropy": 1.3862933665513992,
      "paired_len12_selected_oracle_gap_recovered": NaN,
      "paired_len12_selected_pair_answer_consistency": 1.0,
      "paired_len12_selected_pair_both_correct": 0.015625,
      "paired_len12_oracle_pair_both_correct": 0.015625,
      "paired_len12_selected_pair_state_consistency": 1.0,
      "paired_len12_oracle_pair_state_consistency": 1.0,
      "paired_len24_n": 128.0,
      "paired_len24_beam_count": 4.0,
      "paired_len24_beam0_accuracy": 0.0,
      "paired_len24_selected_accuracy": 0.0,
      "paired_len24_prior_accuracy": 0.0,
      "paired_len24_oracle_accuracy": 0.0,
      "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": 1.0,
      "paired_len24_selector_entropy": 1.386293388903141,
      "paired_len24_selected_oracle_gap_recovered": NaN,
      "paired_len24_selected_pair_answer_consistency": 1.0,
      "paired_len24_selected_pair_both_correct": 0.0,
      "paired_len24_oracle_pair_both_correct": 0.0,
      "paired_len24_selected_pair_state_consistency": 1.0,
      "paired_len24_oracle_pair_state_consistency": 1.0
    },
    {
      "stage": "long",
      "local_step": 1,
      "step": 121,
      "loss": 42.877628326416016,
      "trace_set_loss": 34.18392562866211,
      "executor_loss": 4.57546329498291,
      "state_set_loss": 3.8755736351013184,
      "selector_loss": 1.3850862979888916,
      "diversity_loss": 0.010310644283890724,
      "pair_set_loss": -0.6931470632553101,
      "selector_matches_best_trace": 0.5,
      "best_soft_answer_mass": 0.010301832109689713,
      "mean_soft_answer_mass": 0.01030152291059494,
      "best_target_prob_beam_fraction": 1.0,
      "standard_len12_n": 64.0,
      "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": 1.0,
      "standard_len12_selector_entropy": 1.3862931728363037,
      "standard_len12_selected_oracle_gap_recovered": NaN,
      "standard_len24_n": 64.0,
      "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": 1.03125,
      "standard_len24_selector_entropy": 1.3862931728363037,
      "standard_len24_selected_oracle_gap_recovered": NaN,
      "paraphrase_len12_n": 64.0,
      "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": 0.03125,
      "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": 0.0013020833333333333,
      "paraphrase_len12_oracle_state_prefix_fraction": 0.0013020833333333333,
      "paraphrase_len12_avg_distinct_answers": 1.0,
      "paraphrase_len12_selector_entropy": 1.3862931728363037,
      "paraphrase_len12_selected_oracle_gap_recovered": NaN,
      "paraphrase_len24_n": 64.0,
      "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": 0.0006510416666666666,
      "paraphrase_len24_oracle_state_prefix_fraction": 0.0006510416666666666,
      "paraphrase_len24_avg_distinct_answers": 1.0,
      "paraphrase_len24_selector_entropy": 1.3862931728363037,
      "paraphrase_len24_selected_oracle_gap_recovered": NaN,
      "paired_len12_n": 128.0,
      "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": 0.0,
      "paired_len12_oracle_state_prefix_fraction": 0.0,
      "paired_len12_avg_distinct_answers": 1.0,
      "paired_len12_selector_entropy": 1.3862931728363037,
      "paired_len12_selected_oracle_gap_recovered": NaN,
      "paired_len12_selected_pair_answer_consistency": 1.0,
      "paired_len12_selected_pair_both_correct": 0.015625,
      "paired_len12_oracle_pair_both_correct": 0.015625,
      "paired_len12_selected_pair_state_consistency": 1.0,
      "paired_len12_oracle_pair_state_consistency": 1.0,
      "paired_len24_n": 128.0,
      "paired_len24_beam_count": 4.0,
      "paired_len24_beam0_accuracy": 0.0,
      "paired_len24_selected_accuracy": 0.0,
      "paired_len24_prior_accuracy": 0.0,
      "paired_len24_oracle_accuracy": 0.0,
      "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": 1.0078125,
      "paired_len24_selector_entropy": 1.3862931728363037,
      "paired_len24_selected_oracle_gap_recovered": NaN,
      "paired_len24_selected_pair_answer_consistency": 1.0,
      "paired_len24_selected_pair_both_correct": 0.0,
      "paired_len24_oracle_pair_both_correct": 0.0,
      "paired_len24_selected_pair_state_consistency": 1.0,
      "paired_len24_oracle_pair_state_consistency": 0.984375
    },
    {
      "stage": "long",
      "local_step": 40,
      "step": 160,
      "loss": 39.864715576171875,
      "trace_set_loss": 31.165128707885742,
      "executor_loss": 4.575074195861816,
      "state_set_loss": 3.881743907928467,
      "selector_loss": 1.3855807781219482,
      "diversity_loss": 0.010309449397027493,
      "pair_set_loss": -0.6931471824645996,
      "selector_matches_best_trace": 0.25,
      "best_soft_answer_mass": 0.010305690579116344,
      "mean_soft_answer_mass": 0.010305536910891533,
      "best_target_prob_beam_fraction": 0.5,
      "standard_len12_n": 64.0,
      "standard_len12_beam_count": 4.0,
      "standard_len12_beam0_accuracy": 0.015625,
      "standard_len12_selected_accuracy": 0.015625,
      "standard_len12_prior_accuracy": 0.0,
      "standard_len12_oracle_accuracy": 0.015625,
      "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": 2.84375,
      "standard_len12_selector_entropy": 1.3862876892089844,
      "standard_len12_selected_oracle_gap_recovered": NaN,
      "standard_len24_n": 64.0,
      "standard_len24_beam_count": 4.0,
      "standard_len24_beam0_accuracy": 0.015625,
      "standard_len24_selected_accuracy": 0.015625,
      "standard_len24_prior_accuracy": 0.015625,
      "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": 0.0006510416666666666,
      "standard_len24_oracle_state_prefix_fraction": 0.0006510416666666666,
      "standard_len24_avg_distinct_answers": 2.765625,
      "standard_len24_selector_entropy": 1.3862876892089844,
      "standard_len24_selected_oracle_gap_recovered": 0.0,
      "paraphrase_len12_n": 64.0,
      "paraphrase_len12_beam_count": 4.0,
      "paraphrase_len12_beam0_accuracy": 0.015625,
      "paraphrase_len12_selected_accuracy": 0.015625,
      "paraphrase_len12_prior_accuracy": 0.015625,
      "paraphrase_len12_oracle_accuracy": 0.046875,
      "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": 0.0013020833333333333,
      "paraphrase_len12_oracle_state_prefix_fraction": 0.0013020833333333333,
      "paraphrase_len12_avg_distinct_answers": 2.25,
      "paraphrase_len12_selector_entropy": 1.3862876892089844,
      "paraphrase_len12_selected_oracle_gap_recovered": 0.0,
      "paraphrase_len24_n": 64.0,
      "paraphrase_len24_beam_count": 4.0,
      "paraphrase_len24_beam0_accuracy": 0.0,
      "paraphrase_len24_selected_accuracy": 0.0,
      "paraphrase_len24_prior_accuracy": 0.015625,
      "paraphrase_len24_oracle_accuracy": 0.015625,
      "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": 2.453125,
      "paraphrase_len24_selector_entropy": 1.3862876892089844,
      "paraphrase_len24_selected_oracle_gap_recovered": 0.0,
      "paired_len12_n": 128.0,
      "paired_len12_beam_count": 4.0,
      "paired_len12_beam0_accuracy": 0.0,
      "paired_len12_selected_accuracy": 0.0,
      "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": 0.0026041666666666665,
      "paired_len12_avg_distinct_answers": 2.5078125,
      "paired_len12_selector_entropy": 1.3862876892089844,
      "paired_len12_selected_oracle_gap_recovered": 0.0,
      "paired_len12_selected_pair_answer_consistency": 0.015625,
      "paired_len12_selected_pair_both_correct": 0.0,
      "paired_len12_oracle_pair_both_correct": 0.0,
      "paired_len12_selected_pair_state_consistency": 0.015625,
      "paired_len12_oracle_pair_state_consistency": 0.03125,
      "paired_len24_n": 128.0,
      "paired_len24_beam_count": 4.0,
      "paired_len24_beam0_accuracy": 0.015625,
      "paired_len24_selected_accuracy": 0.015625,
      "paired_len24_prior_accuracy": 0.0078125,
      "paired_len24_oracle_accuracy": 0.0234375,
      "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": 2.6328125,
      "paired_len24_selector_entropy": 1.3862876892089844,
      "paired_len24_selected_oracle_gap_recovered": 0.0,
      "paired_len24_selected_pair_answer_consistency": 0.046875,
      "paired_len24_selected_pair_both_correct": 0.0,
      "paired_len24_oracle_pair_both_correct": 0.0,
      "paired_len24_selected_pair_state_consistency": 0.0,
      "paired_len24_oracle_pair_state_consistency": 0.0
    }
  ],
  "final_metrics": {
    "standard_len12": {
      "n": 64.0,
      "beam_count": 4.0,
      "beam0_accuracy": 0.015625,
      "selected_accuracy": 0.015625,
      "prior_accuracy": 0.0,
      "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": 0.0,
      "oracle_state_prefix_fraction": 0.0,
      "avg_distinct_answers": 2.84375,
      "selector_entropy": 1.3862876892089844,
      "selected_oracle_gap_recovered": NaN
    },
    "standard_len24": {
      "n": 64.0,
      "beam_count": 4.0,
      "beam0_accuracy": 0.015625,
      "selected_accuracy": 0.015625,
      "prior_accuracy": 0.015625,
      "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": 0.0006510416666666666,
      "oracle_state_prefix_fraction": 0.0006510416666666666,
      "avg_distinct_answers": 2.765625,
      "selector_entropy": 1.3862876892089844,
      "selected_oracle_gap_recovered": 0.0
    },
    "paraphrase_len12": {
      "n": 64.0,
      "beam_count": 4.0,
      "beam0_accuracy": 0.015625,
      "selected_accuracy": 0.015625,
      "prior_accuracy": 0.015625,
      "oracle_accuracy": 0.046875,
      "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": 2.25,
      "selector_entropy": 1.3862876892089844,
      "selected_oracle_gap_recovered": 0.0
    },
    "paraphrase_len24": {
      "n": 64.0,
      "beam_count": 4.0,
      "beam0_accuracy": 0.0,
      "selected_accuracy": 0.0,
      "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": 0.0013020833333333333,
      "oracle_state_prefix_fraction": 0.0013020833333333333,
      "avg_distinct_answers": 2.453125,
      "selector_entropy": 1.3862876892089844,
      "selected_oracle_gap_recovered": 0.0
    },
    "paired_len12": {
      "n": 128.0,
      "beam_count": 4.0,
      "beam0_accuracy": 0.0,
      "selected_accuracy": 0.0,
      "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": 0.0026041666666666665,
      "avg_distinct_answers": 2.5078125,
      "selector_entropy": 1.3862876892089844,
      "selected_oracle_gap_recovered": 0.0,
      "selected_pair_answer_consistency": 0.015625,
      "selected_pair_both_correct": 0.0,
      "oracle_pair_both_correct": 0.0,
      "selected_pair_state_consistency": 0.015625,
      "oracle_pair_state_consistency": 0.03125
    },
    "paired_len24": {
      "n": 128.0,
      "beam_count": 4.0,
      "beam0_accuracy": 0.015625,
      "selected_accuracy": 0.015625,
      "prior_accuracy": 0.0078125,
      "oracle_accuracy": 0.0234375,
      "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": 2.6328125,
      "selector_entropy": 1.3862876892089844,
      "selected_oracle_gap_recovered": 0.0,
      "selected_pair_answer_consistency": 0.046875,
      "selected_pair_both_correct": 0.0,
      "oracle_pair_both_correct": 0.0,
      "selected_pair_state_consistency": 0.0,
      "oracle_pair_state_consistency": 0.0
    }
  }
}