{ "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "args": { "run_name": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "model_id": "Qwen/Qwen3-4B", "seed": 456, "modulus": 97, "stage_max_steps": "24,24,24", "stage_min_lengths": "1,1,8", "stage_train_max_lengths": "8,16,24", "stage_steps": "300,150,300", "train_examples": 512, "train_template_mode": "mixed", "eval_lengths": "8,16,24", "eval_examples": 64, "paired_eval_pairs": 32, "batch_size": 8, "grad_accum": 1, "eval_batch_size": 8, "max_length": 2048, "head_width": 512, "compiler_layers": 1, "compiler_heads": 4, "compiler_dropout": 0.05, "expansion_noise": 0.005, "expansion_mode": "copy_last", "trace_loss_weight": 1.0, "executor_loss_weight": 1.0, "state_loss_weight": 1.0, "init_trace_loss_weight": 4.0, "op_trace_loss_weight": 1.0, "arg_trace_loss_weight": 4.0, "direct_head_weight": 0.0, "lr": 0.0002, "weight_decay": 0.0, "optimizer": "paged_adamw_8bit", "max_grad_norm": 1.0, "torch_dtype": "bf16", "load_in_4bit": true, "device_map": "auto", "use_lora": true, "gradient_checkpointing": true, "lora_r": 8, "lora_alpha": 16, "lora_dropout": 0.05, "lora_target_modules": "all-linear", "log_interval": 50, "save_checkpoints": true, "checkpoint_dir": "", "selection_split": "paired_L24", "selection_metric": "executor_pair_both_correct" }, "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, "model_id": "Qwen/Qwen3-4B", "gpu_name": "NVIDIA RTX 6000 Ada Generation", "gpu_vram_gb": 47.383 }, "stages": [ { "name": "stage1_max24", "max_steps": 24, "train_min_len": 1, "train_max_len": 8, "steps": 300 }, { "name": "stage2_max24", "max_steps": 24, "train_min_len": 1, "train_max_len": 16, "steps": 150 }, { "name": "stage3_max24", "max_steps": 24, "train_min_len": 8, "train_max_len": 24, "steps": 300 } ], "train_log": [ { "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "stage": "stage1_max24", "stage_idx": 1, "local_step": 1, "global_step": 1, "compiler_max_steps": 24, "elapsed_sec": 5.143, "expansion_event": "init", "init_loss": 4.572396755218506, "op_loss": 1.1101629734039307, "arg_loss": 4.626572608947754, "state_loss": 4.582656383514404, "state_train_accuracy": 0.0, "executor_loss": 4.58255672454834, "loss": 47.07125473022461 }, { "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "stage": "stage1_max24", "stage_idx": 1, "local_step": 50, "global_step": 50, "compiler_max_steps": 24, "elapsed_sec": 140.017, "expansion_event": "init", "init_loss": 4.490083694458008, "op_loss": 1.0939558744430542, "arg_loss": 3.579193115234375, "state_loss": 4.574100494384766, "state_train_accuracy": 0.04878048598766327, "executor_loss": 4.5643486976623535, "loss": 42.50951385498047 }, { "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "stage": "stage1_max24", "stage_idx": 1, "local_step": 100, "global_step": 100, "compiler_max_steps": 24, "elapsed_sec": 278.078, "expansion_event": "init", "init_loss": 2.8238325119018555, "op_loss": 0.3303409218788147, "arg_loss": 3.3859410285949707, "state_loss": 4.460171222686768, "state_train_accuracy": 0.0, "executor_loss": 4.423162460327148, "loss": 34.05276870727539 }, { "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "stage": "stage1_max24", "stage_idx": 1, "local_step": 150, "global_step": 150, "compiler_max_steps": 24, "elapsed_sec": 416.601, "expansion_event": "init", "init_loss": 0.6055659651756287, "op_loss": 0.2285507470369339, "arg_loss": 0.7770051956176758, "state_loss": 1.805039405822754, "state_train_accuracy": 0.8541666865348816, "executor_loss": 2.6774892807006836, "loss": 10.241364479064941 }, { "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "stage": "stage1_max24", "stage_idx": 1, "local_step": 200, "global_step": 200, "compiler_max_steps": 24, "elapsed_sec": 554.422, "expansion_event": "init", "init_loss": 0.1261308789253235, "op_loss": 0.0003766182635445148, "arg_loss": 0.006923440378159285, "state_loss": 0.11893988400697708, "state_train_accuracy": 1.0, "executor_loss": 0.15252268314361572, "loss": 0.8040564656257629 }, { "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "stage": "stage1_max24", "stage_idx": 1, "local_step": 250, "global_step": 250, "compiler_max_steps": 24, "elapsed_sec": 692.768, "expansion_event": "init", "init_loss": 0.017045512795448303, "op_loss": 1.6939446140895598e-05, "arg_loss": 0.0006808855105191469, "state_loss": 0.02371497079730034, "state_train_accuracy": 1.0, "executor_loss": 0.019661223515868187, "loss": 0.1142987310886383 }, { "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "stage": "stage1_max24", "stage_idx": 1, "local_step": 300, "global_step": 300, "compiler_max_steps": 24, "elapsed_sec": 830.725, "expansion_event": "init", "init_loss": 0.0014189365319907665, "op_loss": 0.6006098985671997, "arg_loss": 0.6670690178871155, "state_loss": 0.8584789037704468, "state_train_accuracy": 0.9318181872367859, "executor_loss": 1.5514957904815674, "loss": 5.684536457061768 }, { "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "stage": "stage2_max24", "stage_idx": 2, "local_step": 1, "global_step": 301, "compiler_max_steps": 24, "elapsed_sec": 963.516, "expansion_event": "same", "init_loss": 0.023859165608882904, "op_loss": 0.946768581867218, "arg_loss": 2.140177011489868, "state_loss": 2.110511064529419, "state_train_accuracy": 0.7611939907073975, "executor_loss": 4.7388811111450195, "loss": 16.45230484008789 }, { "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "stage": "stage2_max24", "stage_idx": 2, "local_step": 50, "global_step": 350, "compiler_max_steps": 24, "elapsed_sec": 1115.022, "expansion_event": "same", "init_loss": 0.0006512956460937858, "op_loss": 1.9381343463464873e-06, "arg_loss": 9.54605420702137e-05, "state_loss": 0.0012202278012409806, "state_train_accuracy": 1.0, "executor_loss": 0.0017348603578284383, "loss": 0.005944050848484039 }, { "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "stage": "stage2_max24", "stage_idx": 2, "local_step": 100, "global_step": 400, "compiler_max_steps": 24, "elapsed_sec": 1269.569, "expansion_event": "same", "init_loss": 0.00026440530200488865, "op_loss": 7.181419641710818e-05, "arg_loss": 7.698447734583169e-05, "state_loss": 0.0010552429594099522, "state_train_accuracy": 1.0, "executor_loss": 0.001454868819564581, "loss": 0.003947485238313675 }, { "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "stage": "stage2_max24", "stage_idx": 2, "local_step": 150, "global_step": 450, "compiler_max_steps": 24, "elapsed_sec": 1424.508, "expansion_event": "same", "init_loss": 0.00029052532045170665, "op_loss": 1.2845811170336674e-06, "arg_loss": 3.93872287531849e-05, "state_loss": 0.0005124824238009751, "state_train_accuracy": 1.0, "executor_loss": 0.0005854535847902298, "loss": 0.0024188708048313856 }, { "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "stage": "stage3_max24", "stage_idx": 3, "local_step": 1, "global_step": 451, "compiler_max_steps": 24, "elapsed_sec": 1557.83, "expansion_event": "same", "init_loss": 0.00017176455003209412, "op_loss": 0.10071111470460892, "arg_loss": 0.3772493302822113, "state_loss": 0.38456979393959045, "state_train_accuracy": 0.961240291595459, "executor_loss": 2.2094902992248535, "loss": 4.204455375671387 }, { "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "stage": "stage3_max24", "stage_idx": 3, "local_step": 50, "global_step": 500, "compiler_max_steps": 24, "elapsed_sec": 1727.701, "expansion_event": "same", "init_loss": 0.0002297356550116092, "op_loss": 3.264350743847899e-05, "arg_loss": 0.08358122408390045, "state_loss": 0.13086549937725067, "state_train_accuracy": 0.9856115579605103, "executor_loss": 1.117319941520691, "loss": 1.583461880683899 }, { "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "stage": "stage3_max24", "stage_idx": 3, "local_step": 100, "global_step": 550, "compiler_max_steps": 24, "elapsed_sec": 1901.198, "expansion_event": "same", "init_loss": 0.00041679892456158996, "op_loss": 1.1920927533992653e-07, "arg_loss": 1.5209135199256707e-05, "state_loss": 0.00041566157597117126, "state_train_accuracy": 1.0, "executor_loss": 0.0006407838664017618, "loss": 0.0027845969889312983 }, { "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "stage": "stage3_max24", "stage_idx": 3, "local_step": 150, "global_step": 600, "compiler_max_steps": 24, "elapsed_sec": 2073.907, "expansion_event": "same", "init_loss": 3.322890552226454e-05, "op_loss": 5.960463766996327e-08, "arg_loss": 9.838423466135282e-06, "state_loss": 0.0001187110465252772, "state_train_accuracy": 1.0, "executor_loss": 0.0001915974571602419, "loss": 0.0004826374351978302 }, { "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "stage": "stage3_max24", "stage_idx": 3, "local_step": 200, "global_step": 650, "compiler_max_steps": 24, "elapsed_sec": 2247.366, "expansion_event": "same", "init_loss": 5.5609278206247836e-05, "op_loss": 9.092959226109087e-05, "arg_loss": 0.10744144022464752, "state_loss": 0.17441460490226746, "state_train_accuracy": 0.9352518320083618, "executor_loss": 0.7770364880561829, "loss": 1.3815302848815918 }, { "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "stage": "stage3_max24", "stage_idx": 3, "local_step": 250, "global_step": 700, "compiler_max_steps": 24, "elapsed_sec": 2420.382, "expansion_event": "same", "init_loss": 0.0006097372970543802, "op_loss": 1.866421189333778e-05, "arg_loss": 0.002024841494858265, "state_loss": 0.01845064014196396, "state_train_accuracy": 1.0, "executor_loss": 0.035603635013103485, "loss": 0.06461125612258911 }, { "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "stage": "stage3_max24", "stage_idx": 3, "local_step": 300, "global_step": 750, "compiler_max_steps": 24, "elapsed_sec": 2593.439, "expansion_event": "same", "init_loss": 0.0005698768072761595, "op_loss": 0.5888500809669495, "arg_loss": 0.8380435705184937, "state_loss": 1.0708882808685303, "state_train_accuracy": 0.8518518209457397, "executor_loss": 1.9390408992767334, "loss": 6.953232765197754 } ], "metrics": [ { "stage": "stage1_max24", "split": "standard_L8", "n": 64, "max_steps": 24, "executor_accuracy": 0.921875, "executor_target_mass": 0.894726037979126, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.98828125, "arg_accuracy": 0.9765625, "program_exact": 0.921875, "state_accuracy": 0.96484375, "state_all_exact": 0.921875, "state_prefix_fraction": 0.96484375, "state_mean_correct_prefix": 7.71875, "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "global_step": 300, "stage_idx": 1 }, { "stage": "stage1_max24", "split": "paraphrase_L8", "n": 64, "max_steps": 24, "executor_accuracy": 0.890625, "executor_target_mass": 0.8773038163781166, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.951171875, "arg_accuracy": 0.943359375, "program_exact": 0.890625, "state_accuracy": 0.9375, "state_all_exact": 0.890625, "state_prefix_fraction": 0.9375, "state_mean_correct_prefix": 7.5, "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "global_step": 300, "stage_idx": 1 }, { "stage": "stage1_max24", "split": "heldout_L8", "n": 64, "max_steps": 24, "executor_accuracy": 0.6875, "executor_target_mass": 0.668368499726057, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.94140625, "arg_accuracy": 0.92578125, "program_exact": 0.6875, "state_accuracy": 0.828125, "state_all_exact": 0.6875, "state_prefix_fraction": 0.828125, "state_mean_correct_prefix": 6.625, "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "global_step": 300, "stage_idx": 1 }, { "stage": "stage1_max24", "split": "paired_L8", "n": 64, "max_steps": 24, "executor_accuracy": 0.859375, "executor_target_mass": 0.8505593836307526, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.96484375, "arg_accuracy": 0.951171875, "program_exact": 0.859375, "state_accuracy": 0.9296875, "state_all_exact": 0.859375, "state_prefix_fraction": 0.9296875, "state_mean_correct_prefix": 7.4375, "executor_pair_answer_consistency": 0.78125, "executor_pair_both_correct": 0.78125, "compiler_pair_program_consistency": 0.78125, "compiler_pair_state_consistency": 0.78125, "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "global_step": 300, "stage_idx": 1 }, { "stage": "stage1_max24", "split": "paired_heldout_L8", "n": 64, "max_steps": 24, "executor_accuracy": 0.875, "executor_target_mass": 0.8671098798513412, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.97265625, "arg_accuracy": 0.958984375, "program_exact": 0.859375, "state_accuracy": 0.93359375, "state_all_exact": 0.859375, "state_prefix_fraction": 0.927734375, "state_mean_correct_prefix": 7.421875, "executor_pair_answer_consistency": 0.75, "executor_pair_both_correct": 0.75, "compiler_pair_program_consistency": 0.71875, "compiler_pair_state_consistency": 0.71875, "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "global_step": 300, "stage_idx": 1 }, { "stage": "stage1_max24", "split": "standard_L16", "n": 64, "max_steps": 24, "executor_accuracy": 0.09375, "executor_target_mass": 0.08063135386328213, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.8505859375, "arg_accuracy": 0.78125, "program_exact": 0.0625, "state_accuracy": 0.69140625, "state_all_exact": 0.0625, "state_prefix_fraction": 0.689453125, "state_mean_correct_prefix": 11.03125, "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "global_step": 300, "stage_idx": 1 }, { "stage": "stage1_max24", "split": "paraphrase_L16", "n": 64, "max_steps": 24, "executor_accuracy": 0.15625, "executor_target_mass": 0.16281192988390103, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.873046875, "arg_accuracy": 0.80859375, "program_exact": 0.15625, "state_accuracy": 0.7587890625, "state_all_exact": 0.15625, "state_prefix_fraction": 0.755859375, "state_mean_correct_prefix": 12.09375, "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "global_step": 300, "stage_idx": 1 }, { "stage": "stage1_max24", "split": "heldout_L16", "n": 64, "max_steps": 24, "executor_accuracy": 0.046875, "executor_target_mass": 0.012676965387072414, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.7119140625, "arg_accuracy": 0.5732421875, "program_exact": 0.0, "state_accuracy": 0.484375, "state_all_exact": 0.0, "state_prefix_fraction": 0.4775390625, "state_mean_correct_prefix": 7.640625, "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "global_step": 300, "stage_idx": 1 }, { "stage": "stage1_max24", "split": "paired_L16", "n": 64, "max_steps": 24, "executor_accuracy": 0.109375, "executor_target_mass": 0.10602510531316511, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.8388671875, "arg_accuracy": 0.7900390625, "program_exact": 0.09375, "state_accuracy": 0.71875, "state_all_exact": 0.09375, "state_prefix_fraction": 0.7158203125, "state_mean_correct_prefix": 11.453125, "executor_pair_answer_consistency": 0.125, "executor_pair_both_correct": 0.0625, "compiler_pair_program_consistency": 0.125, "compiler_pair_state_consistency": 0.125, "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "global_step": 300, "stage_idx": 1 }, { "stage": "stage1_max24", "split": "paired_heldout_L16", "n": 64, "max_steps": 24, "executor_accuracy": 0.09375, "executor_target_mass": 0.0794419888406992, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.80859375, "arg_accuracy": 0.736328125, "program_exact": 0.078125, "state_accuracy": 0.6611328125, "state_all_exact": 0.078125, "state_prefix_fraction": 0.658203125, "state_mean_correct_prefix": 10.53125, "executor_pair_answer_consistency": 0.0, "executor_pair_both_correct": 0.0, "compiler_pair_program_consistency": 0.0, "compiler_pair_state_consistency": 0.0, "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "global_step": 300, "stage_idx": 1 }, { "stage": "stage1_max24", "split": "standard_L24", "n": 64, "max_steps": 24, "executor_accuracy": 0.0, "executor_target_mass": 0.009713857085444033, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.7317708333333334, "arg_accuracy": 0.6041666666666666, "program_exact": 0.0, "state_accuracy": 0.458984375, "state_all_exact": 0.0, "state_prefix_fraction": 0.4537760416666667, "state_mean_correct_prefix": 10.890625, "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "global_step": 300, "stage_idx": 1 }, { "stage": "stage1_max24", "split": "paraphrase_L24", "n": 64, "max_steps": 24, "executor_accuracy": 0.015625, "executor_target_mass": 0.009754908387549222, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.7506510416666666, "arg_accuracy": 0.625, "program_exact": 0.0, "state_accuracy": 0.51171875, "state_all_exact": 0.0, "state_prefix_fraction": 0.5065104166666666, "state_mean_correct_prefix": 12.15625, "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "global_step": 300, "stage_idx": 1 }, { "stage": "stage1_max24", "split": "heldout_L24", "n": 64, "max_steps": 24, "executor_accuracy": 0.015625, "executor_target_mass": 0.010180876590311527, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.6243489583333334, "arg_accuracy": 0.4283854166666667, "program_exact": 0.0, "state_accuracy": 0.3606770833333333, "state_all_exact": 0.0, "state_prefix_fraction": 0.35546875, "state_mean_correct_prefix": 8.53125, "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "global_step": 300, "stage_idx": 1 }, { "stage": "stage1_max24", "split": "paired_L24", "n": 64, "max_steps": 24, "executor_accuracy": 0.0, "executor_target_mass": 0.01051237212959677, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.712890625, "arg_accuracy": 0.5859375, "program_exact": 0.0, "state_accuracy": 0.4752604166666667, "state_all_exact": 0.0, "state_prefix_fraction": 0.4720052083333333, "state_mean_correct_prefix": 11.328125, "executor_pair_answer_consistency": 0.0, "executor_pair_both_correct": 0.0, "compiler_pair_program_consistency": 0.0, "compiler_pair_state_consistency": 0.0, "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "global_step": 300, "stage_idx": 1 }, { "stage": "stage1_max24", "split": "paired_heldout_L24", "n": 64, "max_steps": 24, "executor_accuracy": 0.015625, "executor_target_mass": 0.010014500585384667, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.6438802083333334, "arg_accuracy": 0.46484375, "program_exact": 0.0, "state_accuracy": 0.3873697916666667, "state_all_exact": 0.0, "state_prefix_fraction": 0.3815104166666667, "state_mean_correct_prefix": 9.15625, "executor_pair_answer_consistency": 0.03125, "executor_pair_both_correct": 0.0, "compiler_pair_program_consistency": 0.0, "compiler_pair_state_consistency": 0.0, "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "global_step": 300, "stage_idx": 1 }, { "stage": "stage2_max24", "split": "standard_L8", "n": 64, "max_steps": 24, "executor_accuracy": 1.0, "executor_target_mass": 0.9994774609804153, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 1.0, "arg_accuracy": 1.0, "program_exact": 1.0, "state_accuracy": 1.0, "state_all_exact": 1.0, "state_prefix_fraction": 1.0, "state_mean_correct_prefix": 8.0, "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max24", "split": "paraphrase_L8", "n": 64, "max_steps": 24, "executor_accuracy": 0.984375, "executor_target_mass": 0.9835944026708603, "direct_accuracy": NaN, "init_accuracy": 0.984375, "op_accuracy": 1.0, "arg_accuracy": 1.0, "program_exact": 0.984375, "state_accuracy": 0.984375, "state_all_exact": 0.984375, "state_prefix_fraction": 0.984375, "state_mean_correct_prefix": 7.875, "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max24", "split": "heldout_L8", "n": 64, "max_steps": 24, "executor_accuracy": 0.953125, "executor_target_mass": 0.9528312385082245, "direct_accuracy": NaN, "init_accuracy": 0.984375, "op_accuracy": 0.9921875, "arg_accuracy": 0.98828125, "program_exact": 0.953125, "state_accuracy": 0.958984375, "state_all_exact": 0.953125, "state_prefix_fraction": 0.958984375, "state_mean_correct_prefix": 7.671875, "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max24", "split": "paired_L8", "n": 64, "max_steps": 24, "executor_accuracy": 0.96875, "executor_target_mass": 0.9682839438319206, "direct_accuracy": NaN, "init_accuracy": 0.96875, "op_accuracy": 1.0, "arg_accuracy": 1.0, "program_exact": 0.96875, "state_accuracy": 0.96875, "state_all_exact": 0.96875, "state_prefix_fraction": 0.96875, "state_mean_correct_prefix": 7.75, "executor_pair_answer_consistency": 1.0, "executor_pair_both_correct": 0.96875, "compiler_pair_program_consistency": 1.0, "compiler_pair_state_consistency": 1.0, "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max24", "split": "paired_heldout_L8", "n": 64, "max_steps": 24, "executor_accuracy": 0.953125, "executor_target_mass": 0.9525925442576408, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.98828125, "arg_accuracy": 0.9765625, "program_exact": 0.953125, "state_accuracy": 0.970703125, "state_all_exact": 0.953125, "state_prefix_fraction": 0.970703125, "state_mean_correct_prefix": 7.765625, "executor_pair_answer_consistency": 0.90625, "executor_pair_both_correct": 0.90625, "compiler_pair_program_consistency": 0.90625, "compiler_pair_state_consistency": 0.90625, "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max24", "split": "standard_L16", "n": 64, "max_steps": 24, "executor_accuracy": 1.0, "executor_target_mass": 0.9990692436695099, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 1.0, "arg_accuracy": 1.0, "program_exact": 1.0, "state_accuracy": 1.0, "state_all_exact": 1.0, "state_prefix_fraction": 1.0, "state_mean_correct_prefix": 16.0, "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max24", "split": "paraphrase_L16", "n": 64, "max_steps": 24, "executor_accuracy": 1.0, "executor_target_mass": 0.9982029870152473, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 1.0, "arg_accuracy": 1.0, "program_exact": 1.0, "state_accuracy": 1.0, "state_all_exact": 1.0, "state_prefix_fraction": 1.0, "state_mean_correct_prefix": 16.0, "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max24", "split": "heldout_L16", "n": 64, "max_steps": 24, "executor_accuracy": 0.90625, "executor_target_mass": 0.8963072225451469, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.9619140625, "arg_accuracy": 0.9462890625, "program_exact": 0.890625, "state_accuracy": 0.9345703125, "state_all_exact": 0.890625, "state_prefix_fraction": 0.931640625, "state_mean_correct_prefix": 14.90625, "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max24", "split": "paired_L16", "n": 64, "max_steps": 24, "executor_accuracy": 0.96875, "executor_target_mass": 0.9678082838654518, "direct_accuracy": NaN, "init_accuracy": 0.96875, "op_accuracy": 1.0, "arg_accuracy": 1.0, "program_exact": 0.96875, "state_accuracy": 0.96875, "state_all_exact": 0.96875, "state_prefix_fraction": 0.96875, "state_mean_correct_prefix": 15.5, "executor_pair_answer_consistency": 1.0, "executor_pair_both_correct": 0.96875, "compiler_pair_program_consistency": 1.0, "compiler_pair_state_consistency": 1.0, "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max24", "split": "paired_heldout_L16", "n": 64, "max_steps": 24, "executor_accuracy": 0.8125, "executor_target_mass": 0.8119421228766441, "direct_accuracy": NaN, "init_accuracy": 0.90625, "op_accuracy": 0.97265625, "arg_accuracy": 0.9658203125, "program_exact": 0.8125, "state_accuracy": 0.8447265625, "state_all_exact": 0.8125, "state_prefix_fraction": 0.8447265625, "state_mean_correct_prefix": 13.515625, "executor_pair_answer_consistency": 0.78125, "executor_pair_both_correct": 0.71875, "compiler_pair_program_consistency": 0.78125, "compiler_pair_state_consistency": 0.78125, "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max24", "split": "standard_L24", "n": 64, "max_steps": 24, "executor_accuracy": 0.328125, "executor_target_mass": 0.3394169630482793, "direct_accuracy": NaN, "init_accuracy": 0.96875, "op_accuracy": 0.9700520833333334, "arg_accuracy": 0.94921875, "program_exact": 0.296875, "state_accuracy": 0.8548177083333334, "state_all_exact": 0.296875, "state_prefix_fraction": 0.8509114583333334, "state_mean_correct_prefix": 20.421875, "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max24", "split": "paraphrase_L24", "n": 64, "max_steps": 24, "executor_accuracy": 0.09375, "executor_target_mass": 0.09625981321005384, "direct_accuracy": NaN, "init_accuracy": 0.984375, "op_accuracy": 0.92578125, "arg_accuracy": 0.884765625, "program_exact": 0.078125, "state_accuracy": 0.7942708333333334, "state_all_exact": 0.078125, "state_prefix_fraction": 0.7903645833333334, "state_mean_correct_prefix": 18.96875, "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max24", "split": "heldout_L24", "n": 64, "max_steps": 24, "executor_accuracy": 0.109375, "executor_target_mass": 0.10662691114339395, "direct_accuracy": NaN, "init_accuracy": 0.953125, "op_accuracy": 0.9088541666666666, "arg_accuracy": 0.8528645833333334, "program_exact": 0.109375, "state_accuracy": 0.7506510416666666, "state_all_exact": 0.109375, "state_prefix_fraction": 0.748046875, "state_mean_correct_prefix": 17.953125, "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max24", "split": "paired_L24", "n": 64, "max_steps": 24, "executor_accuracy": 0.21875, "executor_target_mass": 0.21500295028090477, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.947265625, "arg_accuracy": 0.9075520833333334, "program_exact": 0.15625, "state_accuracy": 0.845703125, "state_all_exact": 0.15625, "state_prefix_fraction": 0.8391927083333334, "state_mean_correct_prefix": 20.140625, "executor_pair_answer_consistency": 0.03125, "executor_pair_both_correct": 0.0, "compiler_pair_program_consistency": 0.03125, "compiler_pair_state_consistency": 0.03125, "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max24", "split": "paired_heldout_L24", "n": 64, "max_steps": 24, "executor_accuracy": 0.265625, "executor_target_mass": 0.2353384403977543, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.9544270833333334, "arg_accuracy": 0.9140625, "program_exact": 0.21875, "state_accuracy": 0.828125, "state_all_exact": 0.21875, "state_prefix_fraction": 0.8235677083333334, "state_mean_correct_prefix": 19.765625, "executor_pair_answer_consistency": 0.15625, "executor_pair_both_correct": 0.09375, "compiler_pair_program_consistency": 0.09375, "compiler_pair_state_consistency": 0.09375, "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "global_step": 450, "stage_idx": 2 }, { "stage": "stage3_max24", "split": "standard_L8", "n": 64, "max_steps": 24, "executor_accuracy": 1.0, "executor_target_mass": 0.902509331703186, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 1.0, "arg_accuracy": 1.0, "program_exact": 1.0, "state_accuracy": 1.0, "state_all_exact": 1.0, "state_prefix_fraction": 1.0, "state_mean_correct_prefix": 8.0, "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "paraphrase_L8", "n": 64, "max_steps": 24, "executor_accuracy": 1.0, "executor_target_mass": 0.9052614197134972, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 1.0, "arg_accuracy": 1.0, "program_exact": 1.0, "state_accuracy": 1.0, "state_all_exact": 1.0, "state_prefix_fraction": 1.0, "state_mean_correct_prefix": 8.0, "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "heldout_L8", "n": 64, "max_steps": 24, "executor_accuracy": 1.0, "executor_target_mass": 0.9431563392281532, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 1.0, "arg_accuracy": 1.0, "program_exact": 1.0, "state_accuracy": 1.0, "state_all_exact": 1.0, "state_prefix_fraction": 1.0, "state_mean_correct_prefix": 8.0, "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "paired_L8", "n": 64, "max_steps": 24, "executor_accuracy": 1.0, "executor_target_mass": 0.8710499703884125, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 1.0, "arg_accuracy": 1.0, "program_exact": 1.0, "state_accuracy": 1.0, "state_all_exact": 1.0, "state_prefix_fraction": 1.0, "state_mean_correct_prefix": 8.0, "executor_pair_answer_consistency": 1.0, "executor_pair_both_correct": 1.0, "compiler_pair_program_consistency": 1.0, "compiler_pair_state_consistency": 1.0, "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "paired_heldout_L8", "n": 64, "max_steps": 24, "executor_accuracy": 1.0, "executor_target_mass": 0.9126064330339432, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 1.0, "arg_accuracy": 1.0, "program_exact": 1.0, "state_accuracy": 1.0, "state_all_exact": 1.0, "state_prefix_fraction": 1.0, "state_mean_correct_prefix": 8.0, "executor_pair_answer_consistency": 1.0, "executor_pair_both_correct": 1.0, "compiler_pair_program_consistency": 1.0, "compiler_pair_state_consistency": 1.0, "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "standard_L16", "n": 64, "max_steps": 24, "executor_accuracy": 0.90625, "executor_target_mass": 0.7343185916543007, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.99609375, "arg_accuracy": 0.9931640625, "program_exact": 0.90625, "state_accuracy": 0.9873046875, "state_all_exact": 0.90625, "state_prefix_fraction": 0.9873046875, "state_mean_correct_prefix": 15.796875, "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "paraphrase_L16", "n": 64, "max_steps": 24, "executor_accuracy": 0.96875, "executor_target_mass": 0.8049506023526192, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 1.0, "arg_accuracy": 0.9970703125, "program_exact": 0.96875, "state_accuracy": 0.9853515625, "state_all_exact": 0.96875, "state_prefix_fraction": 0.9853515625, "state_mean_correct_prefix": 15.765625, "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "heldout_L16", "n": 64, "max_steps": 24, "executor_accuracy": 0.875, "executor_target_mass": 0.7134828120470047, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.998046875, "arg_accuracy": 0.9921875, "program_exact": 0.875, "state_accuracy": 0.9833984375, "state_all_exact": 0.875, "state_prefix_fraction": 0.9833984375, "state_mean_correct_prefix": 15.734375, "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "paired_L16", "n": 64, "max_steps": 24, "executor_accuracy": 0.953125, "executor_target_mass": 0.7089060917496681, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 1.0, "arg_accuracy": 0.9970703125, "program_exact": 0.953125, "state_accuracy": 0.9873046875, "state_all_exact": 0.953125, "state_prefix_fraction": 0.9873046875, "state_mean_correct_prefix": 15.796875, "executor_pair_answer_consistency": 0.90625, "executor_pair_both_correct": 0.90625, "compiler_pair_program_consistency": 0.90625, "compiler_pair_state_consistency": 0.90625, "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "paired_heldout_L16", "n": 64, "max_steps": 24, "executor_accuracy": 0.953125, "executor_target_mass": 0.8019792437553406, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.99609375, "arg_accuracy": 0.9931640625, "program_exact": 0.953125, "state_accuracy": 0.9853515625, "state_all_exact": 0.953125, "state_prefix_fraction": 0.9853515625, "state_mean_correct_prefix": 15.765625, "executor_pair_answer_consistency": 0.90625, "executor_pair_both_correct": 0.90625, "compiler_pair_program_consistency": 0.90625, "compiler_pair_state_consistency": 0.90625, "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "standard_L24", "n": 64, "max_steps": 24, "executor_accuracy": 0.015625, "executor_target_mass": 0.014635470899520442, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.9127604166666666, "arg_accuracy": 0.8756510416666666, "program_exact": 0.0, "state_accuracy": 0.720703125, "state_all_exact": 0.0, "state_prefix_fraction": 0.7180989583333334, "state_mean_correct_prefix": 17.234375, "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "paraphrase_L24", "n": 64, "max_steps": 24, "executor_accuracy": 0.015625, "executor_target_mass": 0.00911627832101658, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.8697916666666666, "arg_accuracy": 0.810546875, "program_exact": 0.0, "state_accuracy": 0.7317708333333334, "state_all_exact": 0.0, "state_prefix_fraction": 0.7298177083333334, "state_mean_correct_prefix": 17.515625, "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "heldout_L24", "n": 64, "max_steps": 24, "executor_accuracy": 0.0, "executor_target_mass": 0.0104957745061256, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.8684895833333334, "arg_accuracy": 0.8209635416666666, "program_exact": 0.0, "state_accuracy": 0.7467447916666666, "state_all_exact": 0.0, "state_prefix_fraction": 0.74609375, "state_mean_correct_prefix": 17.90625, "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "paired_L24", "n": 64, "max_steps": 24, "executor_accuracy": 0.0, "executor_target_mass": 0.006589237760636024, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.8997395833333334, "arg_accuracy": 0.8548177083333334, "program_exact": 0.0, "state_accuracy": 0.7421875, "state_all_exact": 0.0, "state_prefix_fraction": 0.7408854166666666, "state_mean_correct_prefix": 17.78125, "executor_pair_answer_consistency": 0.0, "executor_pair_both_correct": 0.0, "compiler_pair_program_consistency": 0.0, "compiler_pair_state_consistency": 0.0, "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "paired_heldout_L24", "n": 64, "max_steps": 24, "executor_accuracy": 0.0, "executor_target_mass": 0.009311080182669684, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.896484375, "arg_accuracy": 0.8548177083333334, "program_exact": 0.0, "state_accuracy": 0.73046875, "state_all_exact": 0.0, "state_prefix_fraction": 0.7298177083333334, "state_mean_correct_prefix": 17.515625, "executor_pair_answer_consistency": 0.0625, "executor_pair_both_correct": 0.0, "compiler_pair_program_consistency": 0.0, "compiler_pair_state_consistency": 0.0, "run": "main_max24_curriculum_seed456", "arm": "max24_curriculum", "seed": 456, "global_step": 750, "stage_idx": 3 } ], "checkpoints": [ { "run": "main_max24_curriculum_seed456", "seed": 456, "tag": "stage1_max24_step00300", "stage": "stage1_max24", "global_step": 300, "compiler_max_steps": 24, "checkpoint_dir": "/workspace/large_artifacts/qwen_compiler_multiseed_reattribution/checkpoints/main_max24_curriculum_seed456/stage1_max24_step00300", "model_checkpoint": "/workspace/large_artifacts/qwen_compiler_multiseed_reattribution/checkpoints/main_max24_curriculum_seed456/stage1_max24_step00300/adapter", "heads_checkpoint": "/workspace/large_artifacts/qwen_compiler_multiseed_reattribution/checkpoints/main_max24_curriculum_seed456/stage1_max24_step00300/heads.pt", "selection_split": "paired_L24", "selection_metric": "executor_pair_both_correct", "selection_value": 0.0, "executor_accuracy": 0.0, "program_exact": 0.0, "state_prefix_fraction": 0.4720052083333333 }, { "run": "main_max24_curriculum_seed456", "seed": 456, "tag": "stage2_max24_step00450", "stage": "stage2_max24", "global_step": 450, "compiler_max_steps": 24, "checkpoint_dir": "/workspace/large_artifacts/qwen_compiler_multiseed_reattribution/checkpoints/main_max24_curriculum_seed456/stage2_max24_step00450", "model_checkpoint": "/workspace/large_artifacts/qwen_compiler_multiseed_reattribution/checkpoints/main_max24_curriculum_seed456/stage2_max24_step00450/adapter", "heads_checkpoint": "/workspace/large_artifacts/qwen_compiler_multiseed_reattribution/checkpoints/main_max24_curriculum_seed456/stage2_max24_step00450/heads.pt", "selection_split": "paired_L24", "selection_metric": "executor_pair_both_correct", "selection_value": 0.0, "executor_accuracy": 0.21875, "program_exact": 0.15625, "state_prefix_fraction": 0.8391927083333334 }, { "run": "main_max24_curriculum_seed456", "seed": 456, "tag": "stage3_max24_step00750", "stage": "stage3_max24", "global_step": 750, "compiler_max_steps": 24, "checkpoint_dir": "/workspace/large_artifacts/qwen_compiler_multiseed_reattribution/checkpoints/main_max24_curriculum_seed456/stage3_max24_step00750", "model_checkpoint": "/workspace/large_artifacts/qwen_compiler_multiseed_reattribution/checkpoints/main_max24_curriculum_seed456/stage3_max24_step00750/adapter", "heads_checkpoint": "/workspace/large_artifacts/qwen_compiler_multiseed_reattribution/checkpoints/main_max24_curriculum_seed456/stage3_max24_step00750/heads.pt", "selection_split": "paired_L24", "selection_metric": "executor_pair_both_correct", "selection_value": 0.0, "executor_accuracy": 0.0, "program_exact": 0.0, "state_prefix_fraction": 0.7408854166666666 } ], "elapsed_sec": 2720.912 }