{ "checkpoints": [ { "adapter_path": "large_artifacts/qwen_onpolicy_repair_compiler/checkpoints/control_gold_only_s256/round01/adapter", "heads_path": "large_artifacts/qwen_onpolicy_repair_compiler/checkpoints/control_gold_only_s256/round01/heads.pt", "round": 1, "run": "control_gold_only_s256" } ], "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": 256, "eval_length": 24, "eval_pairs": 256, "eval_seed": 74001, "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": "control_gold_only_s256", "seed": 74, "state_loss_weight": 0.0, "target_mode": "gold_only", "train_batch_size": 2, "train_examples": 256, "train_length": 24, "train_template_modes": "standard,paraphrase", "val_examples": 128, "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": 188.7753074169159 }, "metrics": { "fresh_paired_len24": { "arg_accuracy": 0.9996744791666666, "arg_pos_accuracy": 0.4248046875, "compiler_pair_program_consistency": 0.984375, "compiler_pair_state_consistency": 0.984375, "direct_accuracy": NaN, "direct_target_mass": NaN, "executor_accuracy": 0.9921875, "executor_pair_answer_consistency": 0.984375, "executor_pair_both_correct": 0.984375, "executor_target_mass": 0.01030929597618524, "init_accuracy": 1.0, "init_pos_accuracy": 1.0, "n": 512.0, "op_accuracy": 1.0, "op_pos_accuracy": 1.0, "pair_true_answer_consistency": 1.0, "pair_true_length_consistency": 1.0, "program_exact": 0.9921875, "repair_avg_candidates": 307.0, "repair_avg_verified_candidates": 2.154296875, "repair_changed_fraction": 0.0078125, "repair_executor_accuracy": 1.0, "repair_found_fraction": 1.0, "repair_pair_answer_consistency": 1.0, "repair_pair_both_correct": 1.0, "repair_pair_program_consistency": 1.0, "repair_pair_state_consistency": 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": 0.9957682291666666, "state_all_exact": 0.9921875, "state_mean_correct_prefix": 23.8984375, "state_prefix_fraction": 0.9957682291666666, "variant": "onpolicy_repair_compiler" }, "fresh_paraphrase_len24": { "arg_accuracy": 0.9988606770833334, "arg_pos_accuracy": 0.4718424479166667, "direct_accuracy": NaN, "direct_target_mass": NaN, "executor_accuracy": 0.97265625, "executor_target_mass": 0.010309280100045726, "init_accuracy": 1.0, "init_pos_accuracy": 1.0, "n": 256.0, "op_accuracy": 1.0, "op_pos_accuracy": 1.0, "program_exact": 0.97265625, "repair_avg_candidates": 307.0, "repair_avg_verified_candidates": 2.1171875, "repair_changed_fraction": 0.02734375, "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": 0.9820963541666666, "state_all_exact": 0.97265625, "state_mean_correct_prefix": 23.5703125, "state_prefix_fraction": 0.9820963541666666, "variant": "onpolicy_repair_compiler" }, "fresh_standard_len24": { "arg_accuracy": 0.9998372395833334, "arg_pos_accuracy": 0.4010416666666667, "direct_accuracy": NaN, "direct_target_mass": NaN, "executor_accuracy": 0.99609375, "executor_target_mass": 0.010309263423550874, "init_accuracy": 1.0, "init_pos_accuracy": 1.0, "n": 256.0, "op_accuracy": 1.0, "op_pos_accuracy": 1.0, "program_exact": 0.99609375, "repair_avg_candidates": 307.0, "repair_avg_verified_candidates": 2.109375, "repair_changed_fraction": 0.00390625, "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": 0.9962565104166666, "state_all_exact": 0.99609375, "state_mean_correct_prefix": 23.91015625, "state_prefix_fraction": 0.9962565104166666, "variant": "onpolicy_repair_compiler" }, "val_len24": { "arg_accuracy": 0.9986979166666666, "arg_pos_accuracy": 0.4449869791666667, "direct_accuracy": NaN, "direct_target_mass": NaN, "executor_accuracy": 0.96875, "executor_target_mass": 0.01030928734689951, "init_accuracy": 1.0, "init_pos_accuracy": 1.0, "n": 128.0, "op_accuracy": 1.0, "op_pos_accuracy": 1.0, "program_exact": 0.96875, "repair_avg_candidates": 307.0, "repair_avg_verified_candidates": 2.4453125, "repair_changed_fraction": 0.03125, "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": 0.9892578125, "state_all_exact": 0.96875, "state_mean_correct_prefix": 23.7421875, "state_prefix_fraction": 0.9892578125, "variant": "onpolicy_repair_compiler" } }, "train_log": [ { "epoch": 0, "fresh_paired_len24_arg_accuracy": 0.9479166666666666, "fresh_paired_len24_arg_pos_accuracy": 0.9481608072916666, "fresh_paired_len24_compiler_pair_program_consistency": 0.72265625, "fresh_paired_len24_compiler_pair_state_consistency": 0.72265625, "fresh_paired_len24_direct_accuracy": NaN, "fresh_paired_len24_direct_target_mass": NaN, "fresh_paired_len24_executor_accuracy": 0.291015625, "fresh_paired_len24_executor_pair_answer_consistency": 0.72265625, "fresh_paired_len24_executor_pair_both_correct": 0.265625, "fresh_paired_len24_executor_target_mass": 0.010309288263670169, "fresh_paired_len24_init_accuracy": 1.0, "fresh_paired_len24_init_pos_accuracy": 1.0, "fresh_paired_len24_op_accuracy": 0.9999186197916666, "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": 0.2890625, "fresh_paired_len24_repair_avg_candidates": 307.0, "fresh_paired_len24_repair_avg_verified_candidates": 1.033203125, "fresh_paired_len24_repair_changed_fraction": 0.353515625, "fresh_paired_len24_repair_executor_accuracy": 0.64453125, "fresh_paired_len24_repair_found_fraction": 0.642578125, "fresh_paired_len24_repair_pair_answer_consistency": 0.82421875, "fresh_paired_len24_repair_pair_both_correct": 0.6171875, "fresh_paired_len24_repair_pair_program_consistency": 0.82421875, "fresh_paired_len24_repair_pair_state_consistency": 0.82421875, "fresh_paired_len24_repair_program_exact": 0.642578125, "fresh_paired_len24_repair_state_accuracy": 0.7451171875, "fresh_paired_len24_repair_state_all_exact": 0.642578125, "fresh_paired_len24_repair_state_mean_correct_prefix": 17.828125, "fresh_paired_len24_repair_state_prefix_fraction": 0.7428385416666666, "fresh_paired_len24_state_accuracy": 0.5570475260416666, "fresh_paired_len24_state_all_exact": 0.2890625, "fresh_paired_len24_state_mean_correct_prefix": 13.314453125, "fresh_paired_len24_state_prefix_fraction": 0.5547688802083334, "fresh_paraphrase_len24_arg_accuracy": 0.9484049479166666, "fresh_paraphrase_len24_arg_pos_accuracy": 0.9498697916666666, "fresh_paraphrase_len24_direct_accuracy": NaN, "fresh_paraphrase_len24_direct_target_mass": NaN, "fresh_paraphrase_len24_executor_accuracy": 0.2890625, "fresh_paraphrase_len24_executor_target_mass": 0.010309269069693983, "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": 0.28125, "fresh_paraphrase_len24_repair_avg_candidates": 307.0, "fresh_paraphrase_len24_repair_avg_verified_candidates": 0.97265625, "fresh_paraphrase_len24_repair_changed_fraction": 0.3828125, "fresh_paraphrase_len24_repair_executor_accuracy": 0.66796875, "fresh_paraphrase_len24_repair_found_fraction": 0.6640625, "fresh_paraphrase_len24_repair_program_exact": 0.6640625, "fresh_paraphrase_len24_repair_state_accuracy": 0.7490234375, "fresh_paraphrase_len24_repair_state_all_exact": 0.6640625, "fresh_paraphrase_len24_repair_state_mean_correct_prefix": 17.9453125, "fresh_paraphrase_len24_repair_state_prefix_fraction": 0.7477213541666666, "fresh_paraphrase_len24_state_accuracy": 0.54248046875, "fresh_paraphrase_len24_state_all_exact": 0.28125, "fresh_paraphrase_len24_state_mean_correct_prefix": 12.91015625, "fresh_paraphrase_len24_state_prefix_fraction": 0.5379231770833334, "fresh_standard_len24_arg_accuracy": 0.94970703125, "fresh_standard_len24_arg_pos_accuracy": 0.9505208333333334, "fresh_standard_len24_direct_accuracy": NaN, "fresh_standard_len24_direct_target_mass": NaN, "fresh_standard_len24_executor_accuracy": 0.28515625, "fresh_standard_len24_executor_target_mass": 0.010309251985745504, "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": 0.28515625, "fresh_standard_len24_repair_avg_candidates": 307.0, "fresh_standard_len24_repair_avg_verified_candidates": 1.05078125, "fresh_standard_len24_repair_changed_fraction": 0.38671875, "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": 0.76953125, "fresh_standard_len24_repair_state_all_exact": 0.671875, "fresh_standard_len24_repair_state_mean_correct_prefix": 18.45703125, "fresh_standard_len24_repair_state_prefix_fraction": 0.76904296875, "fresh_standard_len24_state_accuracy": 0.5870768229166666, "fresh_standard_len24_state_all_exact": 0.28515625, "fresh_standard_len24_state_mean_correct_prefix": 14.078125, "fresh_standard_len24_state_prefix_fraction": 0.5865885416666666, "phase": "baseline", "round": 0, "val_len24_arg_accuracy": 0.947265625, "val_len24_arg_pos_accuracy": 0.9482421875, "val_len24_direct_accuracy": NaN, "val_len24_direct_target_mass": NaN, "val_len24_executor_accuracy": 0.265625, "val_len24_executor_target_mass": 0.010309270990546793, "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": 0.265625, "val_len24_repair_avg_candidates": 307.0, "val_len24_repair_avg_verified_candidates": 1.296875, "val_len24_repair_changed_fraction": 0.4296875, "val_len24_repair_executor_accuracy": 0.6953125, "val_len24_repair_found_fraction": 0.6953125, "val_len24_repair_program_exact": 0.6875, "val_len24_repair_state_accuracy": 0.7666015625, "val_len24_repair_state_all_exact": 0.6953125, "val_len24_repair_state_mean_correct_prefix": 18.3984375, "val_len24_repair_state_prefix_fraction": 0.7666015625, "val_len24_state_accuracy": 0.5436197916666666, "val_len24_state_all_exact": 0.265625, "val_len24_state_mean_correct_prefix": 13.046875, "val_len24_state_prefix_fraction": 0.5436197916666666 }, { "epoch": 1, "fresh_paired_len24_arg_accuracy": 0.9996744791666666, "fresh_paired_len24_arg_pos_accuracy": 0.4248046875, "fresh_paired_len24_compiler_pair_program_consistency": 0.984375, "fresh_paired_len24_compiler_pair_state_consistency": 0.984375, "fresh_paired_len24_direct_accuracy": NaN, "fresh_paired_len24_direct_target_mass": NaN, "fresh_paired_len24_executor_accuracy": 0.9921875, "fresh_paired_len24_executor_pair_answer_consistency": 0.984375, "fresh_paired_len24_executor_pair_both_correct": 0.984375, "fresh_paired_len24_executor_target_mass": 0.01030929597618524, "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": 0.9921875, "fresh_paired_len24_repair_avg_candidates": 307.0, "fresh_paired_len24_repair_avg_verified_candidates": 2.154296875, "fresh_paired_len24_repair_changed_fraction": 0.0078125, "fresh_paired_len24_repair_executor_accuracy": 1.0, "fresh_paired_len24_repair_found_fraction": 1.0, "fresh_paired_len24_repair_pair_answer_consistency": 1.0, "fresh_paired_len24_repair_pair_both_correct": 1.0, "fresh_paired_len24_repair_pair_program_consistency": 1.0, "fresh_paired_len24_repair_pair_state_consistency": 1.0, "fresh_paired_len24_repair_program_exact": 1.0, "fresh_paired_len24_repair_state_accuracy": 1.0, "fresh_paired_len24_repair_state_all_exact": 1.0, "fresh_paired_len24_repair_state_mean_correct_prefix": 24.0, "fresh_paired_len24_repair_state_prefix_fraction": 1.0, "fresh_paired_len24_state_accuracy": 0.9957682291666666, "fresh_paired_len24_state_all_exact": 0.9921875, "fresh_paired_len24_state_mean_correct_prefix": 23.8984375, "fresh_paired_len24_state_prefix_fraction": 0.9957682291666666, "fresh_paraphrase_len24_arg_accuracy": 0.9988606770833334, "fresh_paraphrase_len24_arg_pos_accuracy": 0.4718424479166667, "fresh_paraphrase_len24_direct_accuracy": NaN, "fresh_paraphrase_len24_direct_target_mass": NaN, "fresh_paraphrase_len24_executor_accuracy": 0.97265625, "fresh_paraphrase_len24_executor_target_mass": 0.010309280100045726, "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": 0.97265625, "fresh_paraphrase_len24_repair_avg_candidates": 307.0, "fresh_paraphrase_len24_repair_avg_verified_candidates": 2.1171875, "fresh_paraphrase_len24_repair_changed_fraction": 0.02734375, "fresh_paraphrase_len24_repair_executor_accuracy": 1.0, "fresh_paraphrase_len24_repair_found_fraction": 1.0, "fresh_paraphrase_len24_repair_program_exact": 1.0, "fresh_paraphrase_len24_repair_state_accuracy": 1.0, "fresh_paraphrase_len24_repair_state_all_exact": 1.0, "fresh_paraphrase_len24_repair_state_mean_correct_prefix": 24.0, "fresh_paraphrase_len24_repair_state_prefix_fraction": 1.0, "fresh_paraphrase_len24_state_accuracy": 0.9820963541666666, "fresh_paraphrase_len24_state_all_exact": 0.97265625, "fresh_paraphrase_len24_state_mean_correct_prefix": 23.5703125, "fresh_paraphrase_len24_state_prefix_fraction": 0.9820963541666666, "fresh_standard_len24_arg_accuracy": 0.9998372395833334, "fresh_standard_len24_arg_pos_accuracy": 0.4010416666666667, "fresh_standard_len24_direct_accuracy": NaN, "fresh_standard_len24_direct_target_mass": NaN, "fresh_standard_len24_executor_accuracy": 0.99609375, "fresh_standard_len24_executor_target_mass": 0.010309263423550874, "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": 0.99609375, "fresh_standard_len24_repair_avg_candidates": 307.0, "fresh_standard_len24_repair_avg_verified_candidates": 2.109375, "fresh_standard_len24_repair_changed_fraction": 0.00390625, "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": 0.9962565104166666, "fresh_standard_len24_state_all_exact": 0.99609375, "fresh_standard_len24_state_mean_correct_prefix": 23.91015625, "fresh_standard_len24_state_prefix_fraction": 0.9962565104166666, "phase": "train", "round": 1, "target_active_fraction": 1.0, "target_avg_candidates": 307.0, "target_avg_verified": 1.015625, "target_changed_fraction": 0.3671875, "target_repair_fraction": 0.67578125, "target_source_base_fallback": 0, "target_source_gold": 256, "target_source_gold_fallback": 0, "target_source_repair": 0, "target_source_skipped": 0, "train_executor_loss": 4.574709977954626, "train_gold_trace_loss": 2.5366176776587963, "train_loss": 3.3745813369750977, "train_repair_arg_loss": 0.5471299788914621, "train_repair_init_loss": 7.387337189940224e-05, "train_repair_op_loss": 0.3478022823110223, "train_state_loss": 0.0, "val_len24_arg_accuracy": 0.9986979166666666, "val_len24_arg_pos_accuracy": 0.4449869791666667, "val_len24_direct_accuracy": NaN, "val_len24_direct_target_mass": NaN, "val_len24_executor_accuracy": 0.96875, "val_len24_executor_target_mass": 0.01030928734689951, "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": 0.96875, "val_len24_repair_avg_candidates": 307.0, "val_len24_repair_avg_verified_candidates": 2.4453125, "val_len24_repair_changed_fraction": 0.03125, "val_len24_repair_executor_accuracy": 1.0, "val_len24_repair_found_fraction": 1.0, "val_len24_repair_program_exact": 1.0, "val_len24_repair_state_accuracy": 1.0, "val_len24_repair_state_all_exact": 1.0, "val_len24_repair_state_mean_correct_prefix": 24.0, "val_len24_repair_state_prefix_fraction": 1.0, "val_len24_state_accuracy": 0.9892578125, "val_len24_state_all_exact": 0.96875, "val_len24_state_mean_correct_prefix": 23.7421875, "val_len24_state_prefix_fraction": 0.9892578125 } ] }