{
  "run": "smoke_budgeted_action_value",
  "args": {
    "run_name": "smoke_budgeted_action_value",
    "model_id": "Qwen/Qwen3-4B",
    "load_in_4bit": true,
    "torch_dtype": "bf16",
    "device_map": "auto",
    "gradient_checkpointing": true,
    "lora_r": 8,
    "lora_alpha": 16,
    "lora_dropout": 0.05,
    "lora_target_modules": "all-linear",
    "optimizer": "adamw",
    "lr": 0.0005,
    "weight_decay": 0.0,
    "grad_clip": 1.0,
    "compiler_train_size": 32,
    "verifier_train_size": 32,
    "verifier_collect_limit": 32,
    "verifier_val_collect_limit": 16,
    "val_size": 16,
    "fresh_size": 16,
    "hard_size": 16,
    "max_arith_steps": 4,
    "max_prompt_len": 160,
    "d_model": 96,
    "layers": 1,
    "heads": 4,
    "dropout": 0.1,
    "answer_head_width": 512,
    "batch_size": 16,
    "eval_batch_size": 4,
    "feature_batch_size": 4,
    "compiler_steps": 5,
    "eval_every": 5,
    "quick_eval_size": 16,
    "op_loss_weight": 1.0,
    "arg_loss_weight": 1.0,
    "answer_loss_weight": 0.2,
    "search_topk": 3,
    "max_two_arg_pairs": 8,
    "prefix_collect_beam": 4,
    "prefix_op_topk": 3,
    "prefix_arg_topk": 3,
    "max_negative_prefixes": 24,
    "q_margin_temperature": 4.0,
    "q_positive_loss_weight": 1.0,
    "verifier_d_model": 96,
    "verifier_epochs": 2,
    "verifier_batch_size": 128,
    "verifier_lr": 0.0008,
    "verifier_weight_decay": 0.01,
    "verifier_quick_eval": 1024,
    "max_pos_weight": 8.0,
    "label_modes": [
      "exact",
      "found",
      "qvalue"
    ],
    "beam_width": 6,
    "beam_op_topk": 3,
    "beam_arg_topk": 3,
    "verifier_weights": [
      0.5,
      1.0
    ],
    "seed": 411
  },
  "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,
    "model_id": "Qwen/Qwen3-4B",
    "load_in_4bit": true,
    "lora_r": 8,
    "gpu_name": "NVIDIA RTX 6000 Ada Generation",
    "gpu_vram_gb": 47.383
  },
  "prefix_sample_stats": {
    "train": {
      "source_examples": 32,
      "prefix_samples": 1386,
      "exact_positive_samples": 173,
      "found_positive_samples": 754,
      "q_positive_samples": 754,
      "q_extra_positive_samples": 581,
      "exact_negative_samples": 1213,
      "found_negative_samples": 632,
      "exact_positive_rate": 0.12481962481962482,
      "found_positive_rate": 0.5440115440115441,
      "q_positive_rate": 0.5440115440115441,
      "q_extra_positive_rate": 0.41919191919191917,
      "mean_q_value": 0.24813481460288672,
      "mean_nonzero_q_value": 0.4561204947474814,
      "mean_correct_rank": 3.616710875331565
    },
    "val": {
      "source_examples": 16,
      "prefix_samples": 728,
      "exact_positive_samples": 101,
      "found_positive_samples": 395,
      "q_positive_samples": 395,
      "q_extra_positive_samples": 294,
      "exact_negative_samples": 627,
      "found_negative_samples": 333,
      "exact_positive_rate": 0.13873626373626374,
      "found_positive_rate": 0.5425824175824175,
      "q_positive_rate": 0.5425824175824175,
      "q_extra_positive_rate": 0.40384615384615385,
      "mean_q_value": 0.27251302032722946,
      "mean_nonzero_q_value": 0.5022518450587925,
      "mean_correct_rank": 3.711392405063291
    }
  },
  "label_modes": [
    "exact",
    "found",
    "qvalue"
  ],
  "checkpoint": "large_artifacts/qwen_budgeted_action_value_compiler/checkpoints/smoke_budgeted_action_value",
  "results": [
    {
      "run": "smoke_budgeted_action_value",
      "phase": "final",
      "split": "val_mixed",
      "decoder": "greedy",
      "n": 16,
      "accuracy": 0.0,
      "program_exact": 0.0,
      "valid_rate": 0.0,
      "mean_completed": 1.0,
      "oracle_accuracy": 0.0,
      "mean_expansions": 1.0,
      "mean_score": -59.42477186024189
    },
    {
      "run": "smoke_budgeted_action_value",
      "phase": "final",
      "split": "val_mixed",
      "decoder": "local_answer",
      "n": 16,
      "accuracy": 0.0,
      "program_exact": 0.0,
      "valid_rate": 0.0,
      "mean_completed": 1.0,
      "oracle_accuracy": 0.0,
      "mean_expansions": 173.0,
      "mean_score": -59.42477186024189
    },
    {
      "run": "smoke_budgeted_action_value",
      "phase": "final",
      "split": "val_mixed",
      "decoder": "beam_logprob",
      "n": 16,
      "accuracy": 0.0,
      "program_exact": 0.0,
      "valid_rate": 1.0,
      "mean_completed": 10.625,
      "oracle_accuracy": 0.25,
      "mean_expansions": 93.3125,
      "mean_score": -10.301724545657635
    },
    {
      "run": "smoke_budgeted_action_value",
      "phase": "final",
      "split": "val_mixed",
      "decoder": "beam_exact_w0.5",
      "n": 16,
      "accuracy": 0.0625,
      "program_exact": 0.0,
      "valid_rate": 1.0,
      "mean_completed": 7.875,
      "oracle_accuracy": 0.25,
      "mean_expansions": 67.375,
      "mean_score": -16.377454889938235
    },
    {
      "run": "smoke_budgeted_action_value",
      "phase": "final",
      "split": "val_mixed",
      "decoder": "beam_exact_w1",
      "n": 16,
      "accuracy": 0.0625,
      "program_exact": 0.0,
      "valid_rate": 1.0,
      "mean_completed": 8.0,
      "oracle_accuracy": 0.375,
      "mean_expansions": 73.0,
      "mean_score": -19.691822301596403
    },
    {
      "run": "smoke_budgeted_action_value",
      "phase": "final",
      "split": "val_mixed",
      "decoder": "beam_found_w0.5",
      "n": 16,
      "accuracy": 0.0,
      "program_exact": 0.0,
      "valid_rate": 1.0,
      "mean_completed": 35.5625,
      "oracle_accuracy": 0.3125,
      "mean_expansions": 220.875,
      "mean_score": -12.134162303991616
    },
    {
      "run": "smoke_budgeted_action_value",
      "phase": "final",
      "split": "val_mixed",
      "decoder": "beam_found_w1",
      "n": 16,
      "accuracy": 0.0,
      "program_exact": 0.0,
      "valid_rate": 1.0,
      "mean_completed": 43.1875,
      "oracle_accuracy": 0.375,
      "mean_expansions": 275.125,
      "mean_score": -13.966600062325597
    },
    {
      "run": "smoke_budgeted_action_value",
      "phase": "final",
      "split": "val_mixed",
      "decoder": "beam_qvalue_w0.5",
      "n": 16,
      "accuracy": 0.0,
      "program_exact": 0.0,
      "valid_rate": 1.0,
      "mean_completed": 9.9375,
      "oracle_accuracy": 0.1875,
      "mean_expansions": 104.875,
      "mean_score": -13.754114136099815
    },
    {
      "run": "smoke_budgeted_action_value",
      "phase": "final",
      "split": "val_mixed",
      "decoder": "beam_qvalue_w1",
      "n": 16,
      "accuracy": 0.0,
      "program_exact": 0.0,
      "valid_rate": 1.0,
      "mean_completed": 10.8125,
      "oracle_accuracy": 0.25,
      "mean_expansions": 104.25,
      "mean_score": -16.550101779401302
    },
    {
      "run": "smoke_budgeted_action_value",
      "phase": "final",
      "split": "fresh_standard",
      "decoder": "greedy",
      "n": 16,
      "accuracy": 0.0,
      "program_exact": 0.0,
      "valid_rate": 0.0,
      "mean_completed": 1.0,
      "oracle_accuracy": 0.0,
      "mean_expansions": 1.0,
      "mean_score": -59.39359011501074
    },
    {
      "run": "smoke_budgeted_action_value",
      "phase": "final",
      "split": "fresh_standard",
      "decoder": "local_answer",
      "n": 16,
      "accuracy": 0.0,
      "program_exact": 0.0,
      "valid_rate": 0.0,
      "mean_completed": 1.0,
      "oracle_accuracy": 0.0,
      "mean_expansions": 173.0,
      "mean_score": -59.39359011501074
    },
    {
      "run": "smoke_budgeted_action_value",
      "phase": "final",
      "split": "fresh_standard",
      "decoder": "beam_logprob",
      "n": 16,
      "accuracy": 0.0625,
      "program_exact": 0.0625,
      "valid_rate": 1.0,
      "mean_completed": 10.375,
      "oracle_accuracy": 0.25,
      "mean_expansions": 91.4375,
      "mean_score": -10.314935177564621
    },
    {
      "run": "smoke_budgeted_action_value",
      "phase": "final",
      "split": "fresh_standard",
      "decoder": "beam_exact_w0.5",
      "n": 16,
      "accuracy": 0.0625,
      "program_exact": 0.0,
      "valid_rate": 1.0,
      "mean_completed": 8.0,
      "oracle_accuracy": 0.25,
      "mean_expansions": 68.9375,
      "mean_score": -16.382874958217144
    },
    {
      "run": "smoke_budgeted_action_value",
      "phase": "final",
      "split": "fresh_standard",
      "decoder": "beam_exact_w1",
      "n": 16,
      "accuracy": 0.0625,
      "program_exact": 0.0,
      "valid_rate": 1.0,
      "mean_completed": 8.0,
      "oracle_accuracy": 0.3125,
      "mean_expansions": 73.0,
      "mean_score": -19.699042290449142
    },
    {
      "run": "smoke_budgeted_action_value",
      "phase": "final",
      "split": "fresh_standard",
      "decoder": "beam_found_w0.5",
      "n": 16,
      "accuracy": 0.0625,
      "program_exact": 0.0625,
      "valid_rate": 1.0,
      "mean_completed": 35.8125,
      "oracle_accuracy": 0.25,
      "mean_expansions": 225.8125,
      "mean_score": -12.149026732891798
    },
    {
      "run": "smoke_budgeted_action_value",
      "phase": "final",
      "split": "fresh_standard",
      "decoder": "beam_found_w1",
      "n": 16,
      "accuracy": 0.0625,
      "program_exact": 0.0625,
      "valid_rate": 1.0,
      "mean_completed": 42.3125,
      "oracle_accuracy": 0.375,
      "mean_expansions": 274.5625,
      "mean_score": -13.983118288218975
    },
    {
      "run": "smoke_budgeted_action_value",
      "phase": "final",
      "split": "fresh_standard",
      "decoder": "beam_qvalue_w0.5",
      "n": 16,
      "accuracy": 0.0,
      "program_exact": 0.0,
      "valid_rate": 1.0,
      "mean_completed": 10.5625,
      "oracle_accuracy": 0.1875,
      "mean_expansions": 108.0,
      "mean_score": -13.766476429998875
    },
    {
      "run": "smoke_budgeted_action_value",
      "phase": "final",
      "split": "fresh_standard",
      "decoder": "beam_qvalue_w1",
      "n": 16,
      "accuracy": 0.0,
      "program_exact": 0.0,
      "valid_rate": 1.0,
      "mean_completed": 10.8125,
      "oracle_accuracy": 0.25,
      "mean_expansions": 106.125,
      "mean_score": -16.559976652264595
    },
    {
      "run": "smoke_budgeted_action_value",
      "phase": "final",
      "split": "fresh_paraphrase",
      "decoder": "greedy",
      "n": 16,
      "accuracy": 0.0,
      "program_exact": 0.0,
      "valid_rate": 0.0,
      "mean_completed": 1.0,
      "oracle_accuracy": 0.0,
      "mean_expansions": 1.0,
      "mean_score": -59.3363731354475
    },
    {
      "run": "smoke_budgeted_action_value",
      "phase": "final",
      "split": "fresh_paraphrase",
      "decoder": "local_answer",
      "n": 16,
      "accuracy": 0.0,
      "program_exact": 0.0,
      "valid_rate": 0.0,
      "mean_completed": 1.0,
      "oracle_accuracy": 0.0,
      "mean_expansions": 173.0,
      "mean_score": -59.3363731354475
    },
    {
      "run": "smoke_budgeted_action_value",
      "phase": "final",
      "split": "fresh_paraphrase",
      "decoder": "beam_logprob",
      "n": 16,
      "accuracy": 0.0,
      "program_exact": 0.0,
      "valid_rate": 1.0,
      "mean_completed": 10.3125,
      "oracle_accuracy": 0.3125,
      "mean_expansions": 90.5,
      "mean_score": -10.3016786724329
    },
    {
      "run": "smoke_budgeted_action_value",
      "phase": "final",
      "split": "fresh_paraphrase",
      "decoder": "beam_exact_w0.5",
      "n": 16,
      "accuracy": 0.0625,
      "program_exact": 0.0,
      "valid_rate": 1.0,
      "mean_completed": 8.0,
      "oracle_accuracy": 0.1875,
      "mean_expansions": 68.625,
      "mean_score": -16.39784477232024
    },
    {
      "run": "smoke_budgeted_action_value",
      "phase": "final",
      "split": "fresh_paraphrase",
      "decoder": "beam_exact_w1",
      "n": 16,
      "accuracy": 0.0625,
      "program_exact": 0.0,
      "valid_rate": 1.0,
      "mean_completed": 8.0,
      "oracle_accuracy": 0.1875,
      "mean_expansions": 73.0,
      "mean_score": -19.716248041950166
    },
    {
      "run": "smoke_budgeted_action_value",
      "phase": "final",
      "split": "fresh_paraphrase",
      "decoder": "beam_found_w0.5",
      "n": 16,
      "accuracy": 0.0,
      "program_exact": 0.0,
      "valid_rate": 1.0,
      "mean_completed": 35.1875,
      "oracle_accuracy": 0.3125,
      "mean_expansions": 218.9375,
      "mean_score": -12.156864694319665
    },
    {
      "run": "smoke_budgeted_action_value",
      "phase": "final",
      "split": "fresh_paraphrase",
      "decoder": "beam_found_w1",
      "n": 16,
      "accuracy": 0.0,
      "program_exact": 0.0,
      "valid_rate": 1.0,
      "mean_completed": 44.1875,
      "oracle_accuracy": 0.3125,
      "mean_expansions": 286.875,
      "mean_score": -14.012050716206431
    },
    {
      "run": "smoke_budgeted_action_value",
      "phase": "final",
      "split": "fresh_paraphrase",
      "decoder": "beam_qvalue_w0.5",
      "n": 16,
      "accuracy": 0.0,
      "program_exact": 0.0,
      "valid_rate": 1.0,
      "mean_completed": 9.5,
      "oracle_accuracy": 0.1875,
      "mean_expansions": 104.25,
      "mean_score": -13.729154959321022
    },
    {
      "run": "smoke_budgeted_action_value",
      "phase": "final",
      "split": "fresh_paraphrase",
      "decoder": "beam_qvalue_w1",
      "n": 16,
      "accuracy": 0.0,
      "program_exact": 0.0,
      "valid_rate": 1.0,
      "mean_completed": 10.5,
      "oracle_accuracy": 0.125,
      "mean_expansions": 103.0,
      "mean_score": -16.52631863951683
    },
    {
      "run": "smoke_budgeted_action_value",
      "phase": "final",
      "split": "fresh_paired",
      "decoder": "greedy",
      "n": 16,
      "accuracy": 0.0,
      "program_exact": 0.0,
      "valid_rate": 0.0,
      "mean_completed": 1.0,
      "oracle_accuracy": 0.0,
      "mean_expansions": 1.0,
      "mean_score": -59.12471463158727
    },
    {
      "run": "smoke_budgeted_action_value",
      "phase": "final",
      "split": "fresh_paired",
      "decoder": "local_answer",
      "n": 16,
      "accuracy": 0.0,
      "program_exact": 0.0,
      "valid_rate": 0.0,
      "mean_completed": 1.0,
      "oracle_accuracy": 0.0,
      "mean_expansions": 173.0,
      "mean_score": -59.12471463158727
    },
    {
      "run": "smoke_budgeted_action_value",
      "phase": "final",
      "split": "fresh_paired",
      "decoder": "beam_logprob",
      "n": 16,
      "accuracy": 0.0,
      "program_exact": 0.0,
      "valid_rate": 1.0,
      "mean_completed": 10.5625,
      "oracle_accuracy": 0.125,
      "mean_expansions": 93.625,
      "mean_score": -10.277864642441273
    },
    {
      "run": "smoke_budgeted_action_value",
      "phase": "final",
      "split": "fresh_paired",
      "decoder": "beam_exact_w0.5",
      "n": 16,
      "accuracy": 0.0,
      "program_exact": 0.0,
      "valid_rate": 1.0,
      "mean_completed": 7.875,
      "oracle_accuracy": 0.125,
      "mean_expansions": 68.9375,
      "mean_score": -16.352501769084483
    },
    {
      "run": "smoke_budgeted_action_value",
      "phase": "final",
      "split": "fresh_paired",
      "decoder": "beam_exact_w1",
      "n": 16,
      "accuracy": 0.0,
      "program_exact": 0.0,
      "valid_rate": 1.0,
      "mean_completed": 8.0,
      "oracle_accuracy": 0.375,
      "mean_expansions": 73.0,
      "mean_score": -19.675873422063887
    },
    {
      "run": "smoke_budgeted_action_value",
      "phase": "final",
      "split": "fresh_paired",
      "decoder": "beam_found_w0.5",
      "n": 16,
      "accuracy": 0.0,
      "program_exact": 0.0,
      "valid_rate": 1.0,
      "mean_completed": 35.75,
      "oracle_accuracy": 0.25,
      "mean_expansions": 228.6875,
      "mean_score": -12.119172658771276
    },
    {
      "run": "smoke_budgeted_action_value",
      "phase": "final",
      "split": "fresh_paired",
      "decoder": "beam_found_w1",
      "n": 16,
      "accuracy": 0.0,
      "program_exact": 0.0,
      "valid_rate": 1.0,
      "mean_completed": 44.1875,
      "oracle_accuracy": 0.375,
      "mean_expansions": 288.25,
      "mean_score": -13.96048067510128
    },
    {
      "run": "smoke_budgeted_action_value",
      "phase": "final",
      "split": "fresh_paired",
      "decoder": "beam_qvalue_w0.5",
      "n": 16,
      "accuracy": 0.0,
      "program_exact": 0.0,
      "valid_rate": 1.0,
      "mean_completed": 10.1875,
      "oracle_accuracy": 0.1875,
      "mean_expansions": 108.625,
      "mean_score": -13.739179626107216
    },
    {
      "run": "smoke_budgeted_action_value",
      "phase": "final",
      "split": "fresh_paired",
      "decoder": "beam_qvalue_w1",
      "n": 16,
      "accuracy": 0.0,
      "program_exact": 0.0,
      "valid_rate": 1.0,
      "mean_completed": 10.875,
      "oracle_accuracy": 0.375,
      "mean_expansions": 106.125,
      "mean_score": -16.53704956918955
    },
    {
      "run": "smoke_budgeted_action_value",
      "phase": "final",
      "split": "hard_composition",
      "decoder": "greedy",
      "n": 16,
      "accuracy": 0.0,
      "program_exact": 0.0,
      "valid_rate": 0.0,
      "mean_completed": 1.0,
      "oracle_accuracy": 0.0,
      "mean_expansions": 1.0,
      "mean_score": -59.53870113939047
    },
    {
      "run": "smoke_budgeted_action_value",
      "phase": "final",
      "split": "hard_composition",
      "decoder": "local_answer",
      "n": 16,
      "accuracy": 0.0,
      "program_exact": 0.0,
      "valid_rate": 0.0,
      "mean_completed": 1.0,
      "oracle_accuracy": 0.0,
      "mean_expansions": 173.0,
      "mean_score": -59.53870113939047
    },
    {
      "run": "smoke_budgeted_action_value",
      "phase": "final",
      "split": "hard_composition",
      "decoder": "beam_logprob",
      "n": 16,
      "accuracy": 0.0,
      "program_exact": 0.0,
      "valid_rate": 1.0,
      "mean_completed": 10.4375,
      "oracle_accuracy": 0.125,
      "mean_expansions": 91.4375,
      "mean_score": -10.345708139240742
    },
    {
      "run": "smoke_budgeted_action_value",
      "phase": "final",
      "split": "hard_composition",
      "decoder": "beam_exact_w0.5",
      "n": 16,
      "accuracy": 0.125,
      "program_exact": 0.0,
      "valid_rate": 1.0,
      "mean_completed": 8.0,
      "oracle_accuracy": 0.125,
      "mean_expansions": 68.9375,
      "mean_score": -16.432125786319375
    },
    {
      "run": "smoke_budgeted_action_value",
      "phase": "final",
      "split": "hard_composition",
      "decoder": "beam_exact_w1",
      "n": 16,
      "accuracy": 0.125,
      "program_exact": 0.0,
      "valid_rate": 1.0,
      "mean_completed": 8.0,
      "oracle_accuracy": 0.125,
      "mean_expansions": 73.0,
      "mean_score": -19.74342145398259
    },
    {
      "run": "smoke_budgeted_action_value",
      "phase": "final",
      "split": "hard_composition",
      "decoder": "beam_found_w0.5",
      "n": 16,
      "accuracy": 0.0,
      "program_exact": 0.0,
      "valid_rate": 1.0,
      "mean_completed": 35.3125,
      "oracle_accuracy": 0.125,
      "mean_expansions": 221.4375,
      "mean_score": -12.195161638781428
    },
    {
      "run": "smoke_budgeted_action_value",
      "phase": "final",
      "split": "hard_composition",
      "decoder": "beam_found_w1",
      "n": 16,
      "accuracy": 0.0,
      "program_exact": 0.0,
      "valid_rate": 1.0,
      "mean_completed": 43.9375,
      "oracle_accuracy": 0.125,
      "mean_expansions": 280.4375,
      "mean_score": -14.044615138322115
    },
    {
      "run": "smoke_budgeted_action_value",
      "phase": "final",
      "split": "hard_composition",
      "decoder": "beam_qvalue_w0.5",
      "n": 16,
      "accuracy": 0.0,
      "program_exact": 0.0,
      "valid_rate": 1.0,
      "mean_completed": 10.3125,
      "oracle_accuracy": 0.125,
      "mean_expansions": 106.125,
      "mean_score": -13.769628394395113
    },
    {
      "run": "smoke_budgeted_action_value",
      "phase": "final",
      "split": "hard_composition",
      "decoder": "beam_qvalue_w1",
      "n": 16,
      "accuracy": 0.0,
      "program_exact": 0.0,
      "valid_rate": 1.0,
      "mean_completed": 10.75,
      "oracle_accuracy": 0.0,
      "mean_expansions": 105.5,
      "mean_score": -16.56241561472416
    }
  ]
}