{
  "run": "smoke_prefix_state_verifier",
  "args": {
    "run_name": "smoke_prefix_state_verifier",
    "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": 12,
    "verifier_train_size": 12,
    "verifier_collect_limit": 12,
    "verifier_val_collect_limit": 4,
    "val_size": 4,
    "fresh_size": 4,
    "hard_size": 4,
    "max_arith_steps": 4,
    "max_prompt_len": 160,
    "d_model": 64,
    "layers": 1,
    "heads": 4,
    "dropout": 0.1,
    "answer_head_width": 512,
    "batch_size": 4,
    "eval_batch_size": 2,
    "feature_batch_size": 4,
    "compiler_steps": 2,
    "eval_every": 1,
    "quick_eval_size": 4,
    "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": 2,
    "prefix_arg_topk": 2,
    "max_negative_prefixes": 12,
    "verifier_d_model": 64,
    "verifier_epochs": 1,
    "verifier_batch_size": 32,
    "verifier_lr": 0.0008,
    "verifier_weight_decay": 0.01,
    "verifier_quick_eval": 256,
    "max_pos_weight": 8.0,
    "beam_width": 4,
    "beam_op_topk": 2,
    "beam_arg_topk": 2,
    "verifier_weights": [
      1.0,
      2.0
    ],
    "seed": 111
  },
  "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": 12,
      "prefix_samples": 215,
      "positive_samples": 71,
      "negative_samples": 144,
      "positive_rate": 0.3302325581395349
    },
    "val": {
      "source_examples": 4,
      "prefix_samples": 69,
      "positive_samples": 21,
      "negative_samples": 48,
      "positive_rate": 0.30434782608695654
    }
  },
  "checkpoint": "large_artifacts/qwen_prefix_state_process_verifier/checkpoints/smoke_prefix_state_verifier",
  "results": [
    {
      "run": "smoke_prefix_state_verifier",
      "phase": "final",
      "split": "val_mixed",
      "decoder": "greedy",
      "n": 4,
      "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": -91.46871972084045
    },
    {
      "run": "smoke_prefix_state_verifier",
      "phase": "final",
      "split": "val_mixed",
      "decoder": "local_answer",
      "n": 4,
      "accuracy": 0.0,
      "program_exact": 0.0,
      "valid_rate": 0.0,
      "mean_completed": 1.0,
      "oracle_accuracy": 0.0,
      "mean_expansions": 172.5,
      "mean_score": -91.46871972084045
    },
    {
      "run": "smoke_prefix_state_verifier",
      "phase": "final",
      "split": "val_mixed",
      "decoder": "beam_logprob",
      "n": 4,
      "accuracy": 0.0,
      "program_exact": 0.0,
      "valid_rate": 1.0,
      "mean_completed": 3.0,
      "oracle_accuracy": 0.0,
      "mean_expansions": 84.5,
      "mean_score": -12.096357762813568
    },
    {
      "run": "smoke_prefix_state_verifier",
      "phase": "final",
      "split": "val_mixed",
      "decoder": "beam_verifier_w1",
      "n": 4,
      "accuracy": 0.0,
      "program_exact": 0.0,
      "valid_rate": 1.0,
      "mean_completed": 3.0,
      "oracle_accuracy": 0.0,
      "mean_expansions": 84.5,
      "mean_score": -15.3281908929348
    },
    {
      "run": "smoke_prefix_state_verifier",
      "phase": "final",
      "split": "val_mixed",
      "decoder": "beam_verifier_w2",
      "n": 4,
      "accuracy": 0.0,
      "program_exact": 0.0,
      "valid_rate": 1.0,
      "mean_completed": 3.0,
      "oracle_accuracy": 0.0,
      "mean_expansions": 84.5,
      "mean_score": -18.47849154472351
    },
    {
      "run": "smoke_prefix_state_verifier",
      "phase": "final",
      "split": "fresh_standard",
      "decoder": "greedy",
      "n": 4,
      "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": -92.15719223022461
    },
    {
      "run": "smoke_prefix_state_verifier",
      "phase": "final",
      "split": "fresh_standard",
      "decoder": "local_answer",
      "n": 4,
      "accuracy": 0.25,
      "program_exact": 0.0,
      "valid_rate": 0.25,
      "mean_completed": 1.0,
      "oracle_accuracy": 0.25,
      "mean_expansions": 128.5,
      "mean_score": -88.00021919608116
    },
    {
      "run": "smoke_prefix_state_verifier",
      "phase": "final",
      "split": "fresh_standard",
      "decoder": "beam_logprob",
      "n": 4,
      "accuracy": 0.0,
      "program_exact": 0.0,
      "valid_rate": 1.0,
      "mean_completed": 2.5,
      "oracle_accuracy": 0.0,
      "mean_expansions": 79.5,
      "mean_score": -12.715360701084137
    },
    {
      "run": "smoke_prefix_state_verifier",
      "phase": "final",
      "split": "fresh_standard",
      "decoder": "beam_verifier_w1",
      "n": 4,
      "accuracy": 0.0,
      "program_exact": 0.0,
      "valid_rate": 1.0,
      "mean_completed": 2.5,
      "oracle_accuracy": 0.0,
      "mean_expansions": 79.5,
      "mean_score": -16.287627607584
    },
    {
      "run": "smoke_prefix_state_verifier",
      "phase": "final",
      "split": "fresh_standard",
      "decoder": "beam_verifier_w2",
      "n": 4,
      "accuracy": 0.0,
      "program_exact": 0.0,
      "valid_rate": 1.0,
      "mean_completed": 2.5,
      "oracle_accuracy": 0.0,
      "mean_expansions": 79.5,
      "mean_score": -19.725805580615997
    },
    {
      "run": "smoke_prefix_state_verifier",
      "phase": "final",
      "split": "fresh_paraphrase",
      "decoder": "greedy",
      "n": 4,
      "accuracy": 0.0,
      "program_exact": 0.0,
      "valid_rate": 0.25,
      "mean_completed": 1.0,
      "oracle_accuracy": 0.0,
      "mean_expansions": 1.0,
      "mean_score": -87.82914927601814
    },
    {
      "run": "smoke_prefix_state_verifier",
      "phase": "final",
      "split": "fresh_paraphrase",
      "decoder": "local_answer",
      "n": 4,
      "accuracy": 0.0,
      "program_exact": 0.0,
      "valid_rate": 0.25,
      "mean_completed": 1.0,
      "oracle_accuracy": 0.0,
      "mean_expansions": 103.5,
      "mean_score": -87.82914927601814
    },
    {
      "run": "smoke_prefix_state_verifier",
      "phase": "final",
      "split": "fresh_paraphrase",
      "decoder": "beam_logprob",
      "n": 4,
      "accuracy": 0.0,
      "program_exact": 0.0,
      "valid_rate": 1.0,
      "mean_completed": 4.75,
      "oracle_accuracy": 0.0,
      "mean_expansions": 67.25,
      "mean_score": -11.4519761800766
    },
    {
      "run": "smoke_prefix_state_verifier",
      "phase": "final",
      "split": "fresh_paraphrase",
      "decoder": "beam_verifier_w1",
      "n": 4,
      "accuracy": 0.0,
      "program_exact": 0.0,
      "valid_rate": 1.0,
      "mean_completed": 3.0,
      "oracle_accuracy": 0.0,
      "mean_expansions": 58.5,
      "mean_score": -14.41220235824585
    },
    {
      "run": "smoke_prefix_state_verifier",
      "phase": "final",
      "split": "fresh_paraphrase",
      "decoder": "beam_verifier_w2",
      "n": 4,
      "accuracy": 0.0,
      "program_exact": 0.0,
      "valid_rate": 1.0,
      "mean_completed": 3.5,
      "oracle_accuracy": 0.0,
      "mean_expansions": 62.5,
      "mean_score": -17.302002042531967
    },
    {
      "run": "smoke_prefix_state_verifier",
      "phase": "final",
      "split": "fresh_paired",
      "decoder": "greedy",
      "n": 4,
      "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": -91.52104979753494
    },
    {
      "run": "smoke_prefix_state_verifier",
      "phase": "final",
      "split": "fresh_paired",
      "decoder": "local_answer",
      "n": 4,
      "accuracy": 0.0,
      "program_exact": 0.0,
      "valid_rate": 0.0,
      "mean_completed": 1.0,
      "oracle_accuracy": 0.0,
      "mean_expansions": 145.5,
      "mean_score": -91.52104979753494
    },
    {
      "run": "smoke_prefix_state_verifier",
      "phase": "final",
      "split": "fresh_paired",
      "decoder": "beam_logprob",
      "n": 4,
      "accuracy": 0.0,
      "program_exact": 0.0,
      "valid_rate": 1.0,
      "mean_completed": 3.5,
      "oracle_accuracy": 0.0,
      "mean_expansions": 82.25,
      "mean_score": -12.083861291408539
    },
    {
      "run": "smoke_prefix_state_verifier",
      "phase": "final",
      "split": "fresh_paired",
      "decoder": "beam_verifier_w1",
      "n": 4,
      "accuracy": 0.0,
      "program_exact": 0.0,
      "valid_rate": 1.0,
      "mean_completed": 3.0,
      "oracle_accuracy": 0.0,
      "mean_expansions": 79.0,
      "mean_score": -15.346374809741974
    },
    {
      "run": "smoke_prefix_state_verifier",
      "phase": "final",
      "split": "fresh_paired",
      "decoder": "beam_verifier_w2",
      "n": 4,
      "accuracy": 0.0,
      "program_exact": 0.0,
      "valid_rate": 1.0,
      "mean_completed": 3.0,
      "oracle_accuracy": 0.0,
      "mean_expansions": 79.0,
      "mean_score": -18.504558265209198
    },
    {
      "run": "smoke_prefix_state_verifier",
      "phase": "final",
      "split": "hard_composition",
      "decoder": "greedy",
      "n": 4,
      "accuracy": 0.0,
      "program_exact": 0.0,
      "valid_rate": 0.25,
      "mean_completed": 1.0,
      "oracle_accuracy": 0.0,
      "mean_expansions": 1.0,
      "mean_score": -88.01929786801338
    },
    {
      "run": "smoke_prefix_state_verifier",
      "phase": "final",
      "split": "hard_composition",
      "decoder": "local_answer",
      "n": 4,
      "accuracy": 0.25,
      "program_exact": 0.0,
      "valid_rate": 0.5,
      "mean_completed": 1.0,
      "oracle_accuracy": 0.25,
      "mean_expansions": 119.0,
      "mean_score": -83.63417455554008
    },
    {
      "run": "smoke_prefix_state_verifier",
      "phase": "final",
      "split": "hard_composition",
      "decoder": "beam_logprob",
      "n": 4,
      "accuracy": 0.0,
      "program_exact": 0.0,
      "valid_rate": 1.0,
      "mean_completed": 4.5,
      "oracle_accuracy": 0.0,
      "mean_expansions": 75.0,
      "mean_score": -11.49839049577713
    },
    {
      "run": "smoke_prefix_state_verifier",
      "phase": "final",
      "split": "hard_composition",
      "decoder": "beam_verifier_w1",
      "n": 4,
      "accuracy": 0.0,
      "program_exact": 0.0,
      "valid_rate": 1.0,
      "mean_completed": 3.5,
      "oracle_accuracy": 0.0,
      "mean_expansions": 70.0,
      "mean_score": -14.414888486266136
    },
    {
      "run": "smoke_prefix_state_verifier",
      "phase": "final",
      "split": "hard_composition",
      "decoder": "beam_verifier_w2",
      "n": 4,
      "accuracy": 0.0,
      "program_exact": 0.0,
      "valid_rate": 1.0,
      "mean_completed": 3.75,
      "oracle_accuracy": 0.0,
      "mean_expansions": 72.0,
      "mean_score": -17.295059472322464
    }
  ]
}