{ "run": "main_max24_curriculum_s750", "arm": "max24_curriculum", "args": { "run_name": "main_max24_curriculum_s750", "arm": "max24_curriculum", "model_id": "Qwen/Qwen3-4B", "seed": 123, "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_s750", "arm": "max24_curriculum", "stage": "stage1_max24", "stage_idx": 1, "local_step": 1, "global_step": 1, "compiler_max_steps": 24, "elapsed_sec": 5.146, "expansion_event": "init", "init_loss": 4.647781848907471, "op_loss": 1.1190651655197144, "arg_loss": 4.537424087524414, "state_loss": 4.579545497894287, "state_train_accuracy": 0.0, "executor_loss": 4.579961776733398, "loss": 47.01939392089844 }, { "run": "main_max24_curriculum_s750", "arm": "max24_curriculum", "stage": "stage1_max24", "stage_idx": 1, "local_step": 50, "global_step": 50, "compiler_max_steps": 24, "elapsed_sec": 140.407, "expansion_event": "init", "init_loss": 3.7851691246032715, "op_loss": 1.1145936250686646, "arg_loss": 3.486159086227417, "state_loss": 4.563395023345947, "state_train_accuracy": 0.0, "executor_loss": 4.5507731437683105, "loss": 39.3140754699707 }, { "run": "main_max24_curriculum_s750", "arm": "max24_curriculum", "stage": "stage1_max24", "stage_idx": 1, "local_step": 100, "global_step": 100, "compiler_max_steps": 24, "elapsed_sec": 279.176, "expansion_event": "init", "init_loss": 1.598643183708191, "op_loss": 0.27609574794769287, "arg_loss": 2.247499704360962, "state_loss": 4.1215057373046875, "state_train_accuracy": 0.1621621698141098, "executor_loss": 3.98475980758667, "loss": 23.76693344116211 }, { "run": "main_max24_curriculum_s750", "arm": "max24_curriculum", "stage": "stage1_max24", "stage_idx": 1, "local_step": 150, "global_step": 150, "compiler_max_steps": 24, "elapsed_sec": 417.518, "expansion_event": "init", "init_loss": 0.254672110080719, "op_loss": 0.000269276526523754, "arg_loss": 0.03402327746152878, "state_loss": 0.40248334407806396, "state_train_accuracy": 1.0, "executor_loss": 0.4121854603290558, "loss": 1.969719648361206 }, { "run": "main_max24_curriculum_s750", "arm": "max24_curriculum", "stage": "stage1_max24", "stage_idx": 1, "local_step": 200, "global_step": 200, "compiler_max_steps": 24, "elapsed_sec": 556.14, "expansion_event": "init", "init_loss": 0.07918529212474823, "op_loss": 3.772736818064004e-05, "arg_loss": 0.002915093209594488, "state_loss": 0.06997857242822647, "state_train_accuracy": 0.9999999403953552, "executor_loss": 0.09430437535047531, "loss": 0.49272221326828003 }, { "run": "main_max24_curriculum_s750", "arm": "max24_curriculum", "stage": "stage1_max24", "stage_idx": 1, "local_step": 250, "global_step": 250, "compiler_max_steps": 24, "elapsed_sec": 695.093, "expansion_event": "init", "init_loss": 0.013871273025870323, "op_loss": 6.729882443323731e-06, "arg_loss": 0.00042177041177637875, "state_loss": 0.025254076346755028, "state_train_accuracy": 1.0, "executor_loss": 0.015048881992697716, "loss": 0.097481869161129 }, { "run": "main_max24_curriculum_s750", "arm": "max24_curriculum", "stage": "stage1_max24", "stage_idx": 1, "local_step": 300, "global_step": 300, "compiler_max_steps": 24, "elapsed_sec": 833.511, "expansion_event": "init", "init_loss": 0.0013766196789219975, "op_loss": 2.0121060515521094e-06, "arg_loss": 0.000157144881086424, "state_loss": 0.0019630223978310823, "state_train_accuracy": 1.0, "executor_loss": 0.002032978693023324, "loss": 0.01013307087123394 }, { "run": "main_max24_curriculum_s750", "arm": "max24_curriculum", "stage": "stage2_max24", "stage_idx": 2, "local_step": 1, "global_step": 301, "compiler_max_steps": 24, "elapsed_sec": 966.511, "expansion_event": "same", "init_loss": 0.002677652984857559, "op_loss": 3.0777594020037213e-06, "arg_loss": 0.00014313749852590263, "state_loss": 0.0025298879481852055, "state_train_accuracy": 1.0, "executor_loss": 0.0038839455228298903, "loss": 0.0177000742405653 }, { "run": "main_max24_curriculum_s750", "arm": "max24_curriculum", "stage": "stage2_max24", "stage_idx": 2, "local_step": 50, "global_step": 350, "compiler_max_steps": 24, "elapsed_sec": 1116.497, "expansion_event": "same", "init_loss": 0.004545530304312706, "op_loss": 4.291508957976475e-06, "arg_loss": 0.0007059452473185956, "state_loss": 0.009621313773095608, "state_train_accuracy": 1.0, "executor_loss": 0.00800724420696497, "loss": 0.038638751953840256 }, { "run": "main_max24_curriculum_s750", "arm": "max24_curriculum", "stage": "stage2_max24", "stage_idx": 2, "local_step": 100, "global_step": 400, "compiler_max_steps": 24, "elapsed_sec": 1270.0, "expansion_event": "same", "init_loss": 0.00023585463350173086, "op_loss": 1.6949917380770785e-06, "arg_loss": 5.1405553676886484e-05, "state_loss": 0.0004543387331068516, "state_train_accuracy": 1.0, "executor_loss": 0.0006607172545045614, "loss": 0.0022657918743789196 }, { "run": "main_max24_curriculum_s750", "arm": "max24_curriculum", "stage": "stage2_max24", "stage_idx": 2, "local_step": 150, "global_step": 450, "compiler_max_steps": 24, "elapsed_sec": 1424.565, "expansion_event": "same", "init_loss": 7.502423250116408e-05, "op_loss": 4.543872819340322e-06, "arg_loss": 1.7717526134219952e-05, "state_loss": 0.00017519768152851611, "state_train_accuracy": 1.0, "executor_loss": 0.00021200023184064776, "loss": 0.0007627087761647999 }, { "run": "main_max24_curriculum_s750", "arm": "max24_curriculum", "stage": "stage3_max24", "stage_idx": 3, "local_step": 1, "global_step": 451, "compiler_max_steps": 24, "elapsed_sec": 1558.294, "expansion_event": "same", "init_loss": 0.00013807366485707462, "op_loss": 0.0701206848025322, "arg_loss": 0.3927837014198303, "state_loss": 0.2860243618488312, "state_train_accuracy": 0.9760000705718994, "executor_loss": 1.5218749046325684, "loss": 3.44970703125 }, { "run": "main_max24_curriculum_s750", "arm": "max24_curriculum", "stage": "stage3_max24", "stage_idx": 3, "local_step": 50, "global_step": 500, "compiler_max_steps": 24, "elapsed_sec": 1728.641, "expansion_event": "same", "init_loss": 0.015495509840548038, "op_loss": 0.5880939364433289, "arg_loss": 0.5341256260871887, "state_loss": 1.291869878768921, "state_train_accuracy": 0.8211382031440735, "executor_loss": 3.1044883728027344, "loss": 7.182936668395996 }, { "run": "main_max24_curriculum_s750", "arm": "max24_curriculum", "stage": "stage3_max24", "stage_idx": 3, "local_step": 100, "global_step": 550, "compiler_max_steps": 24, "elapsed_sec": 1902.747, "expansion_event": "same", "init_loss": 0.00021131656831130385, "op_loss": 6.293611659202725e-05, "arg_loss": 0.00044673046795651317, "state_loss": 0.00408415962010622, "state_train_accuracy": 0.9999999403953552, "executor_loss": 0.007983573712408543, "loss": 0.014762856997549534 }, { "run": "main_max24_curriculum_s750", "arm": "max24_curriculum", "stage": "stage3_max24", "stage_idx": 3, "local_step": 150, "global_step": 600, "compiler_max_steps": 24, "elapsed_sec": 2076.611, "expansion_event": "same", "init_loss": 0.0001232430076925084, "op_loss": 4.074378921359312e-06, "arg_loss": 6.780593685107306e-05, "state_loss": 0.0007325169281102717, "state_train_accuracy": 1.0, "executor_loss": 0.0013777215499430895, "loss": 0.0028785085305571556 }, { "run": "main_max24_curriculum_s750", "arm": "max24_curriculum", "stage": "stage3_max24", "stage_idx": 3, "local_step": 200, "global_step": 650, "compiler_max_steps": 24, "elapsed_sec": 2250.237, "expansion_event": "same", "init_loss": 0.00011783969239331782, "op_loss": 1.8096036001225002e-05, "arg_loss": 0.0022954707965254784, "state_loss": 0.014353461563587189, "state_train_accuracy": 1.0, "executor_loss": 0.039732858538627625, "loss": 0.06375765800476074 }, { "run": "main_max24_curriculum_s750", "arm": "max24_curriculum", "stage": "stage3_max24", "stage_idx": 3, "local_step": 250, "global_step": 700, "compiler_max_steps": 24, "elapsed_sec": 2424.48, "expansion_event": "same", "init_loss": 4.8531757784076035e-05, "op_loss": 0.14326860010623932, "arg_loss": 0.26341453194618225, "state_loss": 0.5327871441841125, "state_train_accuracy": 0.9243698120117188, "executor_loss": 0.9313932657241821, "loss": 2.661301374435425 }, { "run": "main_max24_curriculum_s750", "arm": "max24_curriculum", "stage": "stage3_max24", "stage_idx": 3, "local_step": 300, "global_step": 750, "compiler_max_steps": 24, "elapsed_sec": 2598.38, "expansion_event": "same", "init_loss": 2.9533523047575727e-05, "op_loss": 1.3623822496811044e-06, "arg_loss": 1.1276548320893198e-05, "state_loss": 0.00012964772759005427, "state_train_accuracy": 1.0, "executor_loss": 0.00022861387697048485, "loss": 0.0005228642839938402 } ], "metrics": [ { "stage": "stage1_max24", "split": "standard_L8", "n": 64, "max_steps": 24, "executor_accuracy": 1.0, "executor_target_mass": 0.9973209276795387, "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_s750", "arm": "max24_curriculum", "global_step": 300, "stage_idx": 1 }, { "stage": "stage1_max24", "split": "paraphrase_L8", "n": 64, "max_steps": 24, "executor_accuracy": 0.984375, "executor_target_mass": 0.9777919203042984, "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_s750", "arm": "max24_curriculum", "global_step": 300, "stage_idx": 1 }, { "stage": "stage1_max24", "split": "heldout_L8", "n": 64, "max_steps": 24, "executor_accuracy": 0.8125, "executor_target_mass": 0.7906221523880959, "direct_accuracy": NaN, "init_accuracy": 0.96875, "op_accuracy": 0.970703125, "arg_accuracy": 0.986328125, "program_exact": 0.8125, "state_accuracy": 0.876953125, "state_all_exact": 0.8125, "state_prefix_fraction": 0.876953125, "state_mean_correct_prefix": 7.015625, "run": "main_max24_curriculum_s750", "arm": "max24_curriculum", "global_step": 300, "stage_idx": 1 }, { "stage": "stage1_max24", "split": "paired_L8", "n": 64, "max_steps": 24, "executor_accuracy": 1.0, "executor_target_mass": 0.99532151222229, "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_s750", "arm": "max24_curriculum", "global_step": 300, "stage_idx": 1 }, { "stage": "stage1_max24", "split": "paired_heldout_L8", "n": 64, "max_steps": 24, "executor_accuracy": 0.921875, "executor_target_mass": 0.9079104661941528, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.98828125, "arg_accuracy": 0.984375, "program_exact": 0.921875, "state_accuracy": 0.970703125, "state_all_exact": 0.921875, "state_prefix_fraction": 0.970703125, "state_mean_correct_prefix": 7.765625, "executor_pair_answer_consistency": 0.84375, "executor_pair_both_correct": 0.84375, "compiler_pair_program_consistency": 0.84375, "compiler_pair_state_consistency": 0.84375, "run": "main_max24_curriculum_s750", "arm": "max24_curriculum", "global_step": 300, "stage_idx": 1 }, { "stage": "stage1_max24", "split": "standard_L16", "n": 64, "max_steps": 24, "executor_accuracy": 0.5, "executor_target_mass": 0.4913706127554178, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.9716796875, "arg_accuracy": 0.951171875, "program_exact": 0.484375, "state_accuracy": 0.90234375, "state_all_exact": 0.484375, "state_prefix_fraction": 0.8994140625, "state_mean_correct_prefix": 14.390625, "run": "main_max24_curriculum_s750", "arm": "max24_curriculum", "global_step": 300, "stage_idx": 1 }, { "stage": "stage1_max24", "split": "paraphrase_L16", "n": 64, "max_steps": 24, "executor_accuracy": 0.265625, "executor_target_mass": 0.26416005846112967, "direct_accuracy": NaN, "init_accuracy": 0.984375, "op_accuracy": 0.9462890625, "arg_accuracy": 0.896484375, "program_exact": 0.25, "state_accuracy": 0.8212890625, "state_all_exact": 0.25, "state_prefix_fraction": 0.818359375, "state_mean_correct_prefix": 13.09375, "run": "main_max24_curriculum_s750", "arm": "max24_curriculum", "global_step": 300, "stage_idx": 1 }, { "stage": "stage1_max24", "split": "heldout_L16", "n": 64, "max_steps": 24, "executor_accuracy": 0.046875, "executor_target_mass": 0.034806322131771594, "direct_accuracy": NaN, "init_accuracy": 0.984375, "op_accuracy": 0.7939453125, "arg_accuracy": 0.703125, "program_exact": 0.03125, "state_accuracy": 0.556640625, "state_all_exact": 0.03125, "state_prefix_fraction": 0.5537109375, "state_mean_correct_prefix": 8.859375, "run": "main_max24_curriculum_s750", "arm": "max24_curriculum", "global_step": 300, "stage_idx": 1 }, { "stage": "stage1_max24", "split": "paired_L16", "n": 64, "max_steps": 24, "executor_accuracy": 0.5, "executor_target_mass": 0.4783429242670536, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.96484375, "arg_accuracy": 0.9443359375, "program_exact": 0.5, "state_accuracy": 0.892578125, "state_all_exact": 0.5, "state_prefix_fraction": 0.8916015625, "state_mean_correct_prefix": 14.265625, "executor_pair_answer_consistency": 0.3125, "executor_pair_both_correct": 0.3125, "compiler_pair_program_consistency": 0.3125, "compiler_pair_state_consistency": 0.3125, "run": "main_max24_curriculum_s750", "arm": "max24_curriculum", "global_step": 300, "stage_idx": 1 }, { "stage": "stage1_max24", "split": "paired_heldout_L16", "n": 64, "max_steps": 24, "executor_accuracy": 0.28125, "executor_target_mass": 0.2733471589162946, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.8994140625, "arg_accuracy": 0.845703125, "program_exact": 0.28125, "state_accuracy": 0.76953125, "state_all_exact": 0.28125, "state_prefix_fraction": 0.7685546875, "state_mean_correct_prefix": 12.296875, "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_s750", "arm": "max24_curriculum", "global_step": 300, "stage_idx": 1 }, { "stage": "stage1_max24", "split": "standard_L24", "n": 64, "max_steps": 24, "executor_accuracy": 0.046875, "executor_target_mass": 0.02993355953367427, "direct_accuracy": NaN, "init_accuracy": 0.984375, "op_accuracy": 0.8704427083333334, "arg_accuracy": 0.7623697916666666, "program_exact": 0.015625, "state_accuracy": 0.6178385416666666, "state_all_exact": 0.015625, "state_prefix_fraction": 0.6126302083333334, "state_mean_correct_prefix": 14.703125, "run": "main_max24_curriculum_s750", "arm": "max24_curriculum", "global_step": 300, "stage_idx": 1 }, { "stage": "stage1_max24", "split": "paraphrase_L24", "n": 64, "max_steps": 24, "executor_accuracy": 0.015625, "executor_target_mass": 0.010647769915522076, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.8072916666666666, "arg_accuracy": 0.6868489583333334, "program_exact": 0.0, "state_accuracy": 0.5813802083333334, "state_all_exact": 0.0, "state_prefix_fraction": 0.578125, "state_mean_correct_prefix": 13.875, "run": "main_max24_curriculum_s750", "arm": "max24_curriculum", "global_step": 300, "stage_idx": 1 }, { "stage": "stage1_max24", "split": "heldout_L24", "n": 64, "max_steps": 24, "executor_accuracy": 0.0, "executor_target_mass": 0.009615171584300697, "direct_accuracy": NaN, "init_accuracy": 0.96875, "op_accuracy": 0.642578125, "arg_accuracy": 0.484375, "program_exact": 0.0, "state_accuracy": 0.37890625, "state_all_exact": 0.0, "state_prefix_fraction": 0.3717447916666667, "state_mean_correct_prefix": 8.921875, "run": "main_max24_curriculum_s750", "arm": "max24_curriculum", "global_step": 300, "stage_idx": 1 }, { "stage": "stage1_max24", "split": "paired_L24", "n": 64, "max_steps": 24, "executor_accuracy": 0.015625, "executor_target_mass": 0.015680883851018734, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.8196614583333334, "arg_accuracy": 0.7154947916666666, "program_exact": 0.0, "state_accuracy": 0.603515625, "state_all_exact": 0.0, "state_prefix_fraction": 0.5983072916666666, "state_mean_correct_prefix": 14.359375, "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_s750", "arm": "max24_curriculum", "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": 0.01224271469982341, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.75, "arg_accuracy": 0.6009114583333334, "program_exact": 0.0, "state_accuracy": 0.4654947916666667, "state_all_exact": 0.0, "state_prefix_fraction": 0.4602864583333333, "state_mean_correct_prefix": 11.046875, "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_s750", "arm": "max24_curriculum", "global_step": 300, "stage_idx": 1 }, { "stage": "stage2_max24", "split": "standard_L8", "n": 64, "max_steps": 24, "executor_accuracy": 1.0, "executor_target_mass": 0.9992470368742943, "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_s750", "arm": "max24_curriculum", "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max24", "split": "paraphrase_L8", "n": 64, "max_steps": 24, "executor_accuracy": 0.96875, "executor_target_mass": 0.9590947479009628, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.99609375, "arg_accuracy": 1.0, "program_exact": 0.96875, "state_accuracy": 0.978515625, "state_all_exact": 0.96875, "state_prefix_fraction": 0.978515625, "state_mean_correct_prefix": 7.828125, "run": "main_max24_curriculum_s750", "arm": "max24_curriculum", "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max24", "split": "heldout_L8", "n": 64, "max_steps": 24, "executor_accuracy": 0.34375, "executor_target_mass": 0.3237335029989481, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.85546875, "arg_accuracy": 0.966796875, "program_exact": 0.34375, "state_accuracy": 0.599609375, "state_all_exact": 0.34375, "state_prefix_fraction": 0.599609375, "state_mean_correct_prefix": 4.796875, "run": "main_max24_curriculum_s750", "arm": "max24_curriculum", "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max24", "split": "paired_L8", "n": 64, "max_steps": 24, "executor_accuracy": 1.0, "executor_target_mass": 0.9966191723942757, "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_s750", "arm": "max24_curriculum", "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max24", "split": "paired_heldout_L8", "n": 64, "max_steps": 24, "executor_accuracy": 0.609375, "executor_target_mass": 0.6084847375750542, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.92578125, "arg_accuracy": 0.98828125, "program_exact": 0.59375, "state_accuracy": 0.759765625, "state_all_exact": 0.609375, "state_prefix_fraction": 0.759765625, "state_mean_correct_prefix": 6.078125, "executor_pair_answer_consistency": 0.21875, "executor_pair_both_correct": 0.21875, "compiler_pair_program_consistency": 0.1875, "compiler_pair_state_consistency": 0.21875, "run": "main_max24_curriculum_s750", "arm": "max24_curriculum", "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max24", "split": "standard_L16", "n": 64, "max_steps": 24, "executor_accuracy": 1.0, "executor_target_mass": 0.9990184754133224, "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_s750", "arm": "max24_curriculum", "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max24", "split": "paraphrase_L16", "n": 64, "max_steps": 24, "executor_accuracy": 0.96875, "executor_target_mass": 0.9736568331718445, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.9970703125, "arg_accuracy": 1.0, "program_exact": 0.96875, "state_accuracy": 0.986328125, "state_all_exact": 0.96875, "state_prefix_fraction": 0.986328125, "state_mean_correct_prefix": 15.78125, "run": "main_max24_curriculum_s750", "arm": "max24_curriculum", "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max24", "split": "heldout_L16", "n": 64, "max_steps": 24, "executor_accuracy": 0.109375, "executor_target_mass": 0.08366138595829398, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.8837890625, "arg_accuracy": 0.98828125, "program_exact": 0.109375, "state_accuracy": 0.42578125, "state_all_exact": 0.109375, "state_prefix_fraction": 0.4208984375, "state_mean_correct_prefix": 6.734375, "run": "main_max24_curriculum_s750", "arm": "max24_curriculum", "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max24", "split": "paired_L16", "n": 64, "max_steps": 24, "executor_accuracy": 1.0, "executor_target_mass": 0.995872974395752, "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, "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_s750", "arm": "max24_curriculum", "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max24", "split": "paired_heldout_L16", "n": 64, "max_steps": 24, "executor_accuracy": 0.59375, "executor_target_mass": 0.5699768923223019, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.9287109375, "arg_accuracy": 0.9765625, "program_exact": 0.59375, "state_accuracy": 0.720703125, "state_all_exact": 0.59375, "state_prefix_fraction": 0.720703125, "state_mean_correct_prefix": 11.53125, "executor_pair_answer_consistency": 0.1875, "executor_pair_both_correct": 0.1875, "compiler_pair_program_consistency": 0.1875, "compiler_pair_state_consistency": 0.1875, "run": "main_max24_curriculum_s750", "arm": "max24_curriculum", "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max24", "split": "standard_L24", "n": 64, "max_steps": 24, "executor_accuracy": 0.34375, "executor_target_mass": 0.32322241738438606, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.96484375, "arg_accuracy": 0.9427083333333334, "program_exact": 0.34375, "state_accuracy": 0.88671875, "state_all_exact": 0.34375, "state_prefix_fraction": 0.8860677083333334, "state_mean_correct_prefix": 21.265625, "run": "main_max24_curriculum_s750", "arm": "max24_curriculum", "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max24", "split": "paraphrase_L24", "n": 64, "max_steps": 24, "executor_accuracy": 0.0, "executor_target_mass": 0.00036581374911293096, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.8658854166666666, "arg_accuracy": 0.8203125, "program_exact": 0.0, "state_accuracy": 0.6979166666666666, "state_all_exact": 0.0, "state_prefix_fraction": 0.6946614583333334, "state_mean_correct_prefix": 16.671875, "run": "main_max24_curriculum_s750", "arm": "max24_curriculum", "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max24", "split": "heldout_L24", "n": 64, "max_steps": 24, "executor_accuracy": 0.015625, "executor_target_mass": 0.017879252747661667, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.77734375, "arg_accuracy": 0.818359375, "program_exact": 0.0, "state_accuracy": 0.330078125, "state_all_exact": 0.0, "state_prefix_fraction": 0.3248697916666667, "state_mean_correct_prefix": 7.796875, "run": "main_max24_curriculum_s750", "arm": "max24_curriculum", "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max24", "split": "paired_L24", "n": 64, "max_steps": 24, "executor_accuracy": 0.1875, "executor_target_mass": 0.18129401747137308, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.919921875, "arg_accuracy": 0.8704427083333334, "program_exact": 0.171875, "state_accuracy": 0.8014322916666666, "state_all_exact": 0.171875, "state_prefix_fraction": 0.7962239583333334, "state_mean_correct_prefix": 19.109375, "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_s750", "arm": "max24_curriculum", "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max24", "split": "paired_heldout_L24", "n": 64, "max_steps": 24, "executor_accuracy": 0.15625, "executor_target_mass": 0.13844707168755122, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.8880208333333334, "arg_accuracy": 0.88671875, "program_exact": 0.140625, "state_accuracy": 0.603515625, "state_all_exact": 0.140625, "state_prefix_fraction": 0.6002604166666666, "state_mean_correct_prefix": 14.40625, "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_s750", "arm": "max24_curriculum", "global_step": 450, "stage_idx": 2 }, { "stage": "stage3_max24", "split": "standard_L8", "n": 64, "max_steps": 24, "executor_accuracy": 1.0, "executor_target_mass": 0.9998665377497673, "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_s750", "arm": "max24_curriculum", "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "paraphrase_L8", "n": 64, "max_steps": 24, "executor_accuracy": 1.0, "executor_target_mass": 0.9998213052749634, "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_s750", "arm": "max24_curriculum", "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "heldout_L8", "n": 64, "max_steps": 24, "executor_accuracy": 0.90625, "executor_target_mass": 0.9067138433456421, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.984375, "arg_accuracy": 1.0, "program_exact": 0.90625, "state_accuracy": 0.93359375, "state_all_exact": 0.90625, "state_prefix_fraction": 0.93359375, "state_mean_correct_prefix": 7.46875, "run": "main_max24_curriculum_s750", "arm": "max24_curriculum", "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "paired_L8", "n": 64, "max_steps": 24, "executor_accuracy": 1.0, "executor_target_mass": 0.9998569265007973, "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_s750", "arm": "max24_curriculum", "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "paired_heldout_L8", "n": 64, "max_steps": 24, "executor_accuracy": 0.9375, "executor_target_mass": 0.9408805966377258, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.990234375, "arg_accuracy": 1.0, "program_exact": 0.9375, "state_accuracy": 0.97265625, "state_all_exact": 0.9375, "state_prefix_fraction": 0.97265625, "state_mean_correct_prefix": 7.78125, "executor_pair_answer_consistency": 0.875, "executor_pair_both_correct": 0.875, "compiler_pair_program_consistency": 0.875, "compiler_pair_state_consistency": 0.875, "run": "main_max24_curriculum_s750", "arm": "max24_curriculum", "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "standard_L16", "n": 64, "max_steps": 24, "executor_accuracy": 1.0, "executor_target_mass": 0.999757669866085, "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_s750", "arm": "max24_curriculum", "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "paraphrase_L16", "n": 64, "max_steps": 24, "executor_accuracy": 1.0, "executor_target_mass": 0.9997715279459953, "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_s750", "arm": "max24_curriculum", "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "heldout_L16", "n": 64, "max_steps": 24, "executor_accuracy": 0.859375, "executor_target_mass": 0.8539284467697144, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.9833984375, "arg_accuracy": 0.994140625, "program_exact": 0.859375, "state_accuracy": 0.8935546875, "state_all_exact": 0.859375, "state_prefix_fraction": 0.8935546875, "state_mean_correct_prefix": 14.296875, "run": "main_max24_curriculum_s750", "arm": "max24_curriculum", "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "paired_L16", "n": 64, "max_steps": 24, "executor_accuracy": 1.0, "executor_target_mass": 0.999778725206852, "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, "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_s750", "arm": "max24_curriculum", "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "paired_heldout_L16", "n": 64, "max_steps": 24, "executor_accuracy": 0.90625, "executor_target_mass": 0.8964850008487701, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.9873046875, "arg_accuracy": 0.982421875, "program_exact": 0.90625, "state_accuracy": 0.94140625, "state_all_exact": 0.90625, "state_prefix_fraction": 0.94140625, "state_mean_correct_prefix": 15.0625, "executor_pair_answer_consistency": 0.8125, "executor_pair_both_correct": 0.8125, "compiler_pair_program_consistency": 0.8125, "compiler_pair_state_consistency": 0.8125, "run": "main_max24_curriculum_s750", "arm": "max24_curriculum", "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "standard_L24", "n": 64, "max_steps": 24, "executor_accuracy": 0.96875, "executor_target_mass": 0.968639187514782, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.9993489583333334, "arg_accuracy": 0.9986979166666666, "program_exact": 0.96875, "state_accuracy": 0.9973958333333334, "state_all_exact": 0.96875, "state_prefix_fraction": 0.9973958333333334, "state_mean_correct_prefix": 23.9375, "run": "main_max24_curriculum_s750", "arm": "max24_curriculum", "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "paraphrase_L24", "n": 64, "max_steps": 24, "executor_accuracy": 0.8125, "executor_target_mass": 0.8143941015005112, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.9934895833333334, "arg_accuracy": 0.9928385416666666, "program_exact": 0.8125, "state_accuracy": 0.98828125, "state_all_exact": 0.8125, "state_prefix_fraction": 0.98828125, "state_mean_correct_prefix": 23.71875, "run": "main_max24_curriculum_s750", "arm": "max24_curriculum", "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "heldout_L24", "n": 64, "max_steps": 24, "executor_accuracy": 0.46875, "executor_target_mass": 0.44387364014983177, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.9674479166666666, "arg_accuracy": 0.9798177083333334, "program_exact": 0.453125, "state_accuracy": 0.853515625, "state_all_exact": 0.453125, "state_prefix_fraction": 0.8528645833333334, "state_mean_correct_prefix": 20.46875, "run": "main_max24_curriculum_s750", "arm": "max24_curriculum", "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "paired_L24", "n": 64, "max_steps": 24, "executor_accuracy": 0.890625, "executor_target_mass": 0.8891530111432076, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.9967447916666666, "arg_accuracy": 0.9947916666666666, "program_exact": 0.890625, "state_accuracy": 0.9928385416666666, "state_all_exact": 0.890625, "state_prefix_fraction": 0.9928385416666666, "state_mean_correct_prefix": 23.828125, "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_s750", "arm": "max24_curriculum", "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "paired_heldout_L24", "n": 64, "max_steps": 24, "executor_accuracy": 0.78125, "executor_target_mass": 0.7465747520327568, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.98046875, "arg_accuracy": 0.9765625, "program_exact": 0.75, "state_accuracy": 0.912109375, "state_all_exact": 0.75, "state_prefix_fraction": 0.9108072916666666, "state_mean_correct_prefix": 21.859375, "executor_pair_answer_consistency": 0.5625, "executor_pair_both_correct": 0.5625, "compiler_pair_program_consistency": 0.5, "compiler_pair_state_consistency": 0.5, "run": "main_max24_curriculum_s750", "arm": "max24_curriculum", "global_step": 750, "stage_idx": 3 } ], "checkpoints": [ { "run": "main_max24_curriculum_s750", "tag": "stage1_max24_step00300", "stage": "stage1_max24", "global_step": 300, "compiler_max_steps": 24, "checkpoint_dir": "/workspace/large_artifacts/qwen_structural_compiler_attribution_ablation/checkpoints/main_max24_curriculum_s750/stage1_max24_step00300", "model_checkpoint": "/workspace/large_artifacts/qwen_structural_compiler_attribution_ablation/checkpoints/main_max24_curriculum_s750/stage1_max24_step00300/adapter", "heads_checkpoint": "/workspace/large_artifacts/qwen_structural_compiler_attribution_ablation/checkpoints/main_max24_curriculum_s750/stage1_max24_step00300/heads.pt", "selection_split": "paired_L24", "selection_metric": "executor_pair_both_correct", "selection_value": 0.0, "executor_accuracy": 0.015625, "program_exact": 0.0, "state_prefix_fraction": 0.5983072916666666 }, { "run": "main_max24_curriculum_s750", "tag": "stage2_max24_step00450", "stage": "stage2_max24", "global_step": 450, "compiler_max_steps": 24, "checkpoint_dir": "/workspace/large_artifacts/qwen_structural_compiler_attribution_ablation/checkpoints/main_max24_curriculum_s750/stage2_max24_step00450", "model_checkpoint": "/workspace/large_artifacts/qwen_structural_compiler_attribution_ablation/checkpoints/main_max24_curriculum_s750/stage2_max24_step00450/adapter", "heads_checkpoint": "/workspace/large_artifacts/qwen_structural_compiler_attribution_ablation/checkpoints/main_max24_curriculum_s750/stage2_max24_step00450/heads.pt", "selection_split": "paired_L24", "selection_metric": "executor_pair_both_correct", "selection_value": 0.0, "executor_accuracy": 0.1875, "program_exact": 0.171875, "state_prefix_fraction": 0.7962239583333334 }, { "run": "main_max24_curriculum_s750", "tag": "stage3_max24_step00750", "stage": "stage3_max24", "global_step": 750, "compiler_max_steps": 24, "checkpoint_dir": "/workspace/large_artifacts/qwen_structural_compiler_attribution_ablation/checkpoints/main_max24_curriculum_s750/stage3_max24_step00750", "model_checkpoint": "/workspace/large_artifacts/qwen_structural_compiler_attribution_ablation/checkpoints/main_max24_curriculum_s750/stage3_max24_step00750/adapter", "heads_checkpoint": "/workspace/large_artifacts/qwen_structural_compiler_attribution_ablation/checkpoints/main_max24_curriculum_s750/stage3_max24_step00750/heads.pt", "selection_split": "paired_L24", "selection_metric": "executor_pair_both_correct", "selection_value": 0.78125, "executor_accuracy": 0.890625, "program_exact": 0.890625, "state_prefix_fraction": 0.9928385416666666 } ], "elapsed_sec": 2726.002 }