{
  "checkpoints": [
    {
      "adapter_path": "large_artifacts/qwen_onpolicy_repair_compiler/checkpoints/pilot_onpolicy_repair_s96_r1/round01/adapter",
      "heads_path": "large_artifacts/qwen_onpolicy_repair_compiler/checkpoints/pilot_onpolicy_repair_s96_r1/round01/heads.pt",
      "round": 1,
      "run": "pilot_onpolicy_repair_s96_r1"
    }
  ],
  "metadata": {
    "args": {
      "arg_gold_loss_weight": 4.0,
      "arg_repair_loss_weight": 4.0,
      "compiler_checkpoint": "large_artifacts/qwen_onpolicy_repair_compiler/checkpoints/fixed_compiler_step00800",
      "epochs_per_round": 1,
      "eval_batch_size": 8,
      "eval_examples": 64,
      "eval_length": 24,
      "eval_pairs": 64,
      "eval_seed": 73001,
      "executor_loss_weight": 0.1,
      "gold_trace_loss_weight": 0.15,
      "grad_clip": 1.0,
      "gradient_checkpointing": true,
      "init_gold_loss_weight": 4.0,
      "init_repair_loss_weight": 4.0,
      "lr": 5e-05,
      "max_length": 384,
      "max_steps": 24,
      "modulus": 97,
      "onpolicy_rounds": 1,
      "op_gold_loss_weight": 1.0,
      "op_repair_loss_weight": 1.0,
      "qwen_batch_size": 8,
      "repair_eval": true,
      "repair_include_args": true,
      "repair_include_init": true,
      "repair_include_ops": true,
      "repair_max_edits": 2,
      "repair_max_pair_arg_slots": 8,
      "repair_same_step_op_arg": true,
      "repair_topk": 3,
      "repair_trace_loss_weight": 1.0,
      "repair_verifier_mode": "state",
      "run_name": "pilot_onpolicy_repair_s96_r1",
      "seed": 73,
      "state_loss_weight": 0.0,
      "target_mode": "repair_or_gold",
      "train_batch_size": 2,
      "train_examples": 96,
      "train_length": 24,
      "train_template_modes": "standard,paraphrase",
      "val_examples": 64,
      "weight_decay": 0.0
    },
    "compiler_args": {
      "arg_distance_temperature": 2.0,
      "arg_reader_mode": "after_op",
      "arg_window": 8,
      "head_width": 768,
      "max_steps": 24,
      "model_id": "Qwen/Qwen3-4B",
      "modulus": 97,
      "rank_temperature": 1.0
    },
    "cuda_available": true,
    "gpu_name": "NVIDIA RTX 6000 Ada Generation",
    "platform": "Linux-6.8.0-62-generic-x86_64-with-glibc2.39",
    "python": "3.12.3",
    "torch": "2.8.0+cu128",
    "train_seconds": 65.3299310207367
  },
  "metrics": {
    "fresh_paired_len24": {
      "arg_accuracy": 0.9954427083333334,
      "arg_pos_accuracy": 0.6399739583333334,
      "compiler_pair_program_consistency": 0.84375,
      "compiler_pair_state_consistency": 0.859375,
      "direct_accuracy": NaN,
      "direct_target_mass": NaN,
      "executor_accuracy": 0.9140625,
      "executor_pair_answer_consistency": 0.859375,
      "executor_pair_both_correct": 0.84375,
      "executor_target_mass": 0.010309276869520545,
      "init_accuracy": 1.0,
      "init_pos_accuracy": 1.0,
      "n": 128.0,
      "op_accuracy": 1.0,
      "op_pos_accuracy": 1.0,
      "pair_true_answer_consistency": 1.0,
      "pair_true_length_consistency": 1.0,
      "program_exact": 0.90625,
      "repair_avg_candidates": 307.0,
      "repair_avg_verified_candidates": 1.96875,
      "repair_changed_fraction": 0.0703125,
      "repair_executor_accuracy": 0.984375,
      "repair_found_fraction": 0.984375,
      "repair_pair_answer_consistency": 0.96875,
      "repair_pair_both_correct": 0.96875,
      "repair_pair_program_consistency": 0.953125,
      "repair_pair_state_consistency": 0.96875,
      "repair_program_exact": 0.9765625,
      "repair_state_accuracy": 0.9876302083333334,
      "repair_state_all_exact": 0.984375,
      "repair_state_mean_correct_prefix": 23.703125,
      "repair_state_prefix_fraction": 0.9876302083333334,
      "state_accuracy": 0.943359375,
      "state_all_exact": 0.9140625,
      "state_mean_correct_prefix": 22.640625,
      "state_prefix_fraction": 0.943359375,
      "variant": "onpolicy_repair_compiler"
    },
    "fresh_paraphrase_len24": {
      "arg_accuracy": 0.9895833333333334,
      "arg_pos_accuracy": 0.8229166666666666,
      "direct_accuracy": NaN,
      "direct_target_mass": NaN,
      "executor_accuracy": 0.796875,
      "executor_target_mass": 0.010309271048754454,
      "init_accuracy": 1.0,
      "init_pos_accuracy": 1.0,
      "n": 64.0,
      "op_accuracy": 1.0,
      "op_pos_accuracy": 1.0,
      "program_exact": 0.796875,
      "repair_avg_candidates": 307.0,
      "repair_avg_verified_candidates": 1.828125,
      "repair_changed_fraction": 0.15625,
      "repair_executor_accuracy": 0.953125,
      "repair_found_fraction": 0.953125,
      "repair_program_exact": 0.953125,
      "repair_state_accuracy": 0.9674479166666666,
      "repair_state_all_exact": 0.953125,
      "repair_state_mean_correct_prefix": 23.21875,
      "repair_state_prefix_fraction": 0.9674479166666666,
      "state_accuracy": 0.89453125,
      "state_all_exact": 0.796875,
      "state_mean_correct_prefix": 21.46875,
      "state_prefix_fraction": 0.89453125,
      "variant": "onpolicy_repair_compiler"
    },
    "fresh_standard_len24": {
      "arg_accuracy": 0.9967447916666666,
      "arg_pos_accuracy": 0.4908854166666667,
      "direct_accuracy": NaN,
      "direct_target_mass": NaN,
      "executor_accuracy": 0.921875,
      "executor_target_mass": 0.010309290722943842,
      "init_accuracy": 1.0,
      "init_pos_accuracy": 1.0,
      "n": 64.0,
      "op_accuracy": 1.0,
      "op_pos_accuracy": 1.0,
      "program_exact": 0.921875,
      "repair_avg_candidates": 307.0,
      "repair_avg_verified_candidates": 1.90625,
      "repair_changed_fraction": 0.078125,
      "repair_executor_accuracy": 1.0,
      "repair_found_fraction": 1.0,
      "repair_program_exact": 1.0,
      "repair_state_accuracy": 1.0,
      "repair_state_all_exact": 1.0,
      "repair_state_mean_correct_prefix": 24.0,
      "repair_state_prefix_fraction": 1.0,
      "state_accuracy": 0.9459635416666666,
      "state_all_exact": 0.921875,
      "state_mean_correct_prefix": 22.703125,
      "state_prefix_fraction": 0.9459635416666666,
      "variant": "onpolicy_repair_compiler"
    },
    "val_len24": {
      "arg_accuracy": 0.9954427083333334,
      "arg_pos_accuracy": 0.65234375,
      "direct_accuracy": NaN,
      "direct_target_mass": NaN,
      "executor_accuracy": 0.90625,
      "executor_target_mass": 0.01030926697421819,
      "init_accuracy": 1.0,
      "init_pos_accuracy": 1.0,
      "n": 64.0,
      "op_accuracy": 1.0,
      "op_pos_accuracy": 1.0,
      "program_exact": 0.90625,
      "repair_avg_candidates": 307.0,
      "repair_avg_verified_candidates": 2.125,
      "repair_changed_fraction": 0.078125,
      "repair_executor_accuracy": 0.984375,
      "repair_found_fraction": 0.984375,
      "repair_program_exact": 0.984375,
      "repair_state_accuracy": 0.9889322916666666,
      "repair_state_all_exact": 0.984375,
      "repair_state_mean_correct_prefix": 23.734375,
      "repair_state_prefix_fraction": 0.9889322916666666,
      "state_accuracy": 0.9459635416666666,
      "state_all_exact": 0.90625,
      "state_mean_correct_prefix": 22.703125,
      "state_prefix_fraction": 0.9459635416666666,
      "variant": "onpolicy_repair_compiler"
    }
  },
  "train_log": [
    {
      "epoch": 0,
      "fresh_paired_len24_arg_accuracy": 0.9462890625,
      "fresh_paired_len24_arg_pos_accuracy": 0.9459635416666666,
      "fresh_paired_len24_compiler_pair_program_consistency": 0.71875,
      "fresh_paired_len24_compiler_pair_state_consistency": 0.71875,
      "fresh_paired_len24_direct_accuracy": NaN,
      "fresh_paired_len24_direct_target_mass": NaN,
      "fresh_paired_len24_executor_accuracy": 0.21875,
      "fresh_paired_len24_executor_pair_answer_consistency": 0.71875,
      "fresh_paired_len24_executor_pair_both_correct": 0.1875,
      "fresh_paired_len24_executor_target_mass": 0.010309272562153637,
      "fresh_paired_len24_init_accuracy": 1.0,
      "fresh_paired_len24_init_pos_accuracy": 1.0,
      "fresh_paired_len24_op_accuracy": 0.9996744791666666,
      "fresh_paired_len24_op_pos_accuracy": 1.0,
      "fresh_paired_len24_pair_true_answer_consistency": 1.0,
      "fresh_paired_len24_pair_true_length_consistency": 1.0,
      "fresh_paired_len24_program_exact": 0.2109375,
      "fresh_paired_len24_repair_avg_candidates": 307.0,
      "fresh_paired_len24_repair_avg_verified_candidates": 0.8984375,
      "fresh_paired_len24_repair_changed_fraction": 0.4765625,
      "fresh_paired_len24_repair_executor_accuracy": 0.6953125,
      "fresh_paired_len24_repair_found_fraction": 0.6875,
      "fresh_paired_len24_repair_pair_answer_consistency": 0.8125,
      "fresh_paired_len24_repair_pair_both_correct": 0.65625,
      "fresh_paired_len24_repair_pair_program_consistency": 0.8125,
      "fresh_paired_len24_repair_pair_state_consistency": 0.8125,
      "fresh_paired_len24_repair_program_exact": 0.6875,
      "fresh_paired_len24_repair_state_accuracy": 0.7805989583333334,
      "fresh_paired_len24_repair_state_all_exact": 0.6875,
      "fresh_paired_len24_repair_state_mean_correct_prefix": 18.609375,
      "fresh_paired_len24_repair_state_prefix_fraction": 0.775390625,
      "fresh_paired_len24_state_accuracy": 0.4918619791666667,
      "fresh_paired_len24_state_all_exact": 0.2109375,
      "fresh_paired_len24_state_mean_correct_prefix": 11.6796875,
      "fresh_paired_len24_state_prefix_fraction": 0.4866536458333333,
      "fresh_paraphrase_len24_arg_accuracy": 0.947265625,
      "fresh_paraphrase_len24_arg_pos_accuracy": 0.9485677083333334,
      "fresh_paraphrase_len24_direct_accuracy": NaN,
      "fresh_paraphrase_len24_direct_target_mass": NaN,
      "fresh_paraphrase_len24_executor_accuracy": 0.3125,
      "fresh_paraphrase_len24_executor_target_mass": 0.010309265344403684,
      "fresh_paraphrase_len24_init_accuracy": 1.0,
      "fresh_paraphrase_len24_init_pos_accuracy": 1.0,
      "fresh_paraphrase_len24_op_accuracy": 1.0,
      "fresh_paraphrase_len24_op_pos_accuracy": 1.0,
      "fresh_paraphrase_len24_program_exact": 0.3125,
      "fresh_paraphrase_len24_repair_avg_candidates": 307.0,
      "fresh_paraphrase_len24_repair_avg_verified_candidates": 0.890625,
      "fresh_paraphrase_len24_repair_changed_fraction": 0.28125,
      "fresh_paraphrase_len24_repair_executor_accuracy": 0.59375,
      "fresh_paraphrase_len24_repair_found_fraction": 0.59375,
      "fresh_paraphrase_len24_repair_program_exact": 0.59375,
      "fresh_paraphrase_len24_repair_state_accuracy": 0.671875,
      "fresh_paraphrase_len24_repair_state_all_exact": 0.59375,
      "fresh_paraphrase_len24_repair_state_mean_correct_prefix": 16.125,
      "fresh_paraphrase_len24_repair_state_prefix_fraction": 0.671875,
      "fresh_paraphrase_len24_state_accuracy": 0.513671875,
      "fresh_paraphrase_len24_state_all_exact": 0.3125,
      "fresh_paraphrase_len24_state_mean_correct_prefix": 12.328125,
      "fresh_paraphrase_len24_state_prefix_fraction": 0.513671875,
      "fresh_standard_len24_arg_accuracy": 0.9505208333333334,
      "fresh_standard_len24_arg_pos_accuracy": 0.9518229166666666,
      "fresh_standard_len24_direct_accuracy": NaN,
      "fresh_standard_len24_direct_target_mass": NaN,
      "fresh_standard_len24_executor_accuracy": 0.296875,
      "fresh_standard_len24_executor_target_mass": 0.010309247532859445,
      "fresh_standard_len24_init_accuracy": 1.0,
      "fresh_standard_len24_init_pos_accuracy": 1.0,
      "fresh_standard_len24_op_accuracy": 1.0,
      "fresh_standard_len24_op_pos_accuracy": 1.0,
      "fresh_standard_len24_program_exact": 0.296875,
      "fresh_standard_len24_repair_avg_candidates": 307.0,
      "fresh_standard_len24_repair_avg_verified_candidates": 1.109375,
      "fresh_standard_len24_repair_changed_fraction": 0.375,
      "fresh_standard_len24_repair_executor_accuracy": 0.671875,
      "fresh_standard_len24_repair_found_fraction": 0.671875,
      "fresh_standard_len24_repair_program_exact": 0.671875,
      "fresh_standard_len24_repair_state_accuracy": 0.7708333333333334,
      "fresh_standard_len24_repair_state_all_exact": 0.671875,
      "fresh_standard_len24_repair_state_mean_correct_prefix": 18.5,
      "fresh_standard_len24_repair_state_prefix_fraction": 0.7708333333333334,
      "fresh_standard_len24_state_accuracy": 0.5735677083333334,
      "fresh_standard_len24_state_all_exact": 0.296875,
      "fresh_standard_len24_state_mean_correct_prefix": 13.765625,
      "fresh_standard_len24_state_prefix_fraction": 0.5735677083333334,
      "phase": "baseline",
      "round": 0,
      "val_len24_arg_accuracy": 0.9498697916666666,
      "val_len24_arg_pos_accuracy": 0.94921875,
      "val_len24_direct_accuracy": NaN,
      "val_len24_direct_target_mass": NaN,
      "val_len24_executor_accuracy": 0.328125,
      "val_len24_executor_target_mass": 0.010309247532859445,
      "val_len24_init_accuracy": 1.0,
      "val_len24_init_pos_accuracy": 1.0,
      "val_len24_op_accuracy": 1.0,
      "val_len24_op_pos_accuracy": 1.0,
      "val_len24_program_exact": 0.328125,
      "val_len24_repair_avg_candidates": 307.0,
      "val_len24_repair_avg_verified_candidates": 1.140625,
      "val_len24_repair_changed_fraction": 0.359375,
      "val_len24_repair_executor_accuracy": 0.6875,
      "val_len24_repair_found_fraction": 0.6875,
      "val_len24_repair_program_exact": 0.6875,
      "val_len24_repair_state_accuracy": 0.7688802083333334,
      "val_len24_repair_state_all_exact": 0.6875,
      "val_len24_repair_state_mean_correct_prefix": 18.453125,
      "val_len24_repair_state_prefix_fraction": 0.7688802083333334,
      "val_len24_state_accuracy": 0.5572916666666666,
      "val_len24_state_all_exact": 0.328125,
      "val_len24_state_mean_correct_prefix": 13.375,
      "val_len24_state_prefix_fraction": 0.5572916666666666
    },
    {
      "epoch": 1,
      "fresh_paired_len24_arg_accuracy": 0.9954427083333334,
      "fresh_paired_len24_arg_pos_accuracy": 0.6399739583333334,
      "fresh_paired_len24_compiler_pair_program_consistency": 0.84375,
      "fresh_paired_len24_compiler_pair_state_consistency": 0.859375,
      "fresh_paired_len24_direct_accuracy": NaN,
      "fresh_paired_len24_direct_target_mass": NaN,
      "fresh_paired_len24_executor_accuracy": 0.9140625,
      "fresh_paired_len24_executor_pair_answer_consistency": 0.859375,
      "fresh_paired_len24_executor_pair_both_correct": 0.84375,
      "fresh_paired_len24_executor_target_mass": 0.010309276869520545,
      "fresh_paired_len24_init_accuracy": 1.0,
      "fresh_paired_len24_init_pos_accuracy": 1.0,
      "fresh_paired_len24_op_accuracy": 1.0,
      "fresh_paired_len24_op_pos_accuracy": 1.0,
      "fresh_paired_len24_pair_true_answer_consistency": 1.0,
      "fresh_paired_len24_pair_true_length_consistency": 1.0,
      "fresh_paired_len24_program_exact": 0.90625,
      "fresh_paired_len24_repair_avg_candidates": 307.0,
      "fresh_paired_len24_repair_avg_verified_candidates": 1.96875,
      "fresh_paired_len24_repair_changed_fraction": 0.0703125,
      "fresh_paired_len24_repair_executor_accuracy": 0.984375,
      "fresh_paired_len24_repair_found_fraction": 0.984375,
      "fresh_paired_len24_repair_pair_answer_consistency": 0.96875,
      "fresh_paired_len24_repair_pair_both_correct": 0.96875,
      "fresh_paired_len24_repair_pair_program_consistency": 0.953125,
      "fresh_paired_len24_repair_pair_state_consistency": 0.96875,
      "fresh_paired_len24_repair_program_exact": 0.9765625,
      "fresh_paired_len24_repair_state_accuracy": 0.9876302083333334,
      "fresh_paired_len24_repair_state_all_exact": 0.984375,
      "fresh_paired_len24_repair_state_mean_correct_prefix": 23.703125,
      "fresh_paired_len24_repair_state_prefix_fraction": 0.9876302083333334,
      "fresh_paired_len24_state_accuracy": 0.943359375,
      "fresh_paired_len24_state_all_exact": 0.9140625,
      "fresh_paired_len24_state_mean_correct_prefix": 22.640625,
      "fresh_paired_len24_state_prefix_fraction": 0.943359375,
      "fresh_paraphrase_len24_arg_accuracy": 0.9895833333333334,
      "fresh_paraphrase_len24_arg_pos_accuracy": 0.8229166666666666,
      "fresh_paraphrase_len24_direct_accuracy": NaN,
      "fresh_paraphrase_len24_direct_target_mass": NaN,
      "fresh_paraphrase_len24_executor_accuracy": 0.796875,
      "fresh_paraphrase_len24_executor_target_mass": 0.010309271048754454,
      "fresh_paraphrase_len24_init_accuracy": 1.0,
      "fresh_paraphrase_len24_init_pos_accuracy": 1.0,
      "fresh_paraphrase_len24_op_accuracy": 1.0,
      "fresh_paraphrase_len24_op_pos_accuracy": 1.0,
      "fresh_paraphrase_len24_program_exact": 0.796875,
      "fresh_paraphrase_len24_repair_avg_candidates": 307.0,
      "fresh_paraphrase_len24_repair_avg_verified_candidates": 1.828125,
      "fresh_paraphrase_len24_repair_changed_fraction": 0.15625,
      "fresh_paraphrase_len24_repair_executor_accuracy": 0.953125,
      "fresh_paraphrase_len24_repair_found_fraction": 0.953125,
      "fresh_paraphrase_len24_repair_program_exact": 0.953125,
      "fresh_paraphrase_len24_repair_state_accuracy": 0.9674479166666666,
      "fresh_paraphrase_len24_repair_state_all_exact": 0.953125,
      "fresh_paraphrase_len24_repair_state_mean_correct_prefix": 23.21875,
      "fresh_paraphrase_len24_repair_state_prefix_fraction": 0.9674479166666666,
      "fresh_paraphrase_len24_state_accuracy": 0.89453125,
      "fresh_paraphrase_len24_state_all_exact": 0.796875,
      "fresh_paraphrase_len24_state_mean_correct_prefix": 21.46875,
      "fresh_paraphrase_len24_state_prefix_fraction": 0.89453125,
      "fresh_standard_len24_arg_accuracy": 0.9967447916666666,
      "fresh_standard_len24_arg_pos_accuracy": 0.4908854166666667,
      "fresh_standard_len24_direct_accuracy": NaN,
      "fresh_standard_len24_direct_target_mass": NaN,
      "fresh_standard_len24_executor_accuracy": 0.921875,
      "fresh_standard_len24_executor_target_mass": 0.010309290722943842,
      "fresh_standard_len24_init_accuracy": 1.0,
      "fresh_standard_len24_init_pos_accuracy": 1.0,
      "fresh_standard_len24_op_accuracy": 1.0,
      "fresh_standard_len24_op_pos_accuracy": 1.0,
      "fresh_standard_len24_program_exact": 0.921875,
      "fresh_standard_len24_repair_avg_candidates": 307.0,
      "fresh_standard_len24_repair_avg_verified_candidates": 1.90625,
      "fresh_standard_len24_repair_changed_fraction": 0.078125,
      "fresh_standard_len24_repair_executor_accuracy": 1.0,
      "fresh_standard_len24_repair_found_fraction": 1.0,
      "fresh_standard_len24_repair_program_exact": 1.0,
      "fresh_standard_len24_repair_state_accuracy": 1.0,
      "fresh_standard_len24_repair_state_all_exact": 1.0,
      "fresh_standard_len24_repair_state_mean_correct_prefix": 24.0,
      "fresh_standard_len24_repair_state_prefix_fraction": 1.0,
      "fresh_standard_len24_state_accuracy": 0.9459635416666666,
      "fresh_standard_len24_state_all_exact": 0.921875,
      "fresh_standard_len24_state_mean_correct_prefix": 22.703125,
      "fresh_standard_len24_state_prefix_fraction": 0.9459635416666666,
      "phase": "train",
      "round": 1,
      "target_active_fraction": 1.0,
      "target_avg_candidates": 307.0,
      "target_avg_verified": 1.0,
      "target_changed_fraction": 0.375,
      "target_repair_fraction": 0.6770833134651184,
      "target_source_base_fallback": 0,
      "target_source_gold_fallback": 31,
      "target_source_repair": 65,
      "target_source_skipped": 0,
      "train_executor_loss": 4.5747106571992235,
      "train_gold_trace_loss": 2.5832231491804123,
      "train_loss": 3.428177679578463,
      "train_repair_arg_loss": 0.5574850688378016,
      "train_repair_init_loss": 0.0002070710528023767,
      "train_repair_op_loss": 0.35245457167426747,
      "train_state_loss": 0.0,
      "val_len24_arg_accuracy": 0.9954427083333334,
      "val_len24_arg_pos_accuracy": 0.65234375,
      "val_len24_direct_accuracy": NaN,
      "val_len24_direct_target_mass": NaN,
      "val_len24_executor_accuracy": 0.90625,
      "val_len24_executor_target_mass": 0.01030926697421819,
      "val_len24_init_accuracy": 1.0,
      "val_len24_init_pos_accuracy": 1.0,
      "val_len24_op_accuracy": 1.0,
      "val_len24_op_pos_accuracy": 1.0,
      "val_len24_program_exact": 0.90625,
      "val_len24_repair_avg_candidates": 307.0,
      "val_len24_repair_avg_verified_candidates": 2.125,
      "val_len24_repair_changed_fraction": 0.078125,
      "val_len24_repair_executor_accuracy": 0.984375,
      "val_len24_repair_found_fraction": 0.984375,
      "val_len24_repair_program_exact": 0.984375,
      "val_len24_repair_state_accuracy": 0.9889322916666666,
      "val_len24_repair_state_all_exact": 0.984375,
      "val_len24_repair_state_mean_correct_prefix": 23.734375,
      "val_len24_repair_state_prefix_fraction": 0.9889322916666666,
      "val_len24_state_accuracy": 0.9459635416666666,
      "val_len24_state_all_exact": 0.90625,
      "val_len24_state_mean_correct_prefix": 22.703125,
      "val_len24_state_prefix_fraction": 0.9459635416666666
    }
  ]
}