{ "run": "main_max24_curriculum_seed789", "arm": "max24_curriculum", "seed": 789, "args": { "run_name": "main_max24_curriculum_seed789", "arm": "max24_curriculum", "model_id": "Qwen/Qwen3-4B", "seed": 789, "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_seed789", "arm": "max24_curriculum", "seed": 789, "stage": "stage1_max24", "stage_idx": 1, "local_step": 1, "global_step": 1, "compiler_max_steps": 24, "elapsed_sec": 5.043, "expansion_event": "init", "init_loss": 4.568448066711426, "op_loss": 1.1028714179992676, "arg_loss": 4.554978847503662, "state_loss": 4.570505142211914, "state_train_accuracy": 0.03125, "executor_loss": 4.579878330230713, "loss": 46.74696350097656 }, { "run": "main_max24_curriculum_seed789", "arm": "max24_curriculum", "seed": 789, "stage": "stage1_max24", "stage_idx": 1, "local_step": 50, "global_step": 50, "compiler_max_steps": 24, "elapsed_sec": 139.893, "expansion_event": "init", "init_loss": 3.2730069160461426, "op_loss": 0.9206094741821289, "arg_loss": 3.592787265777588, "state_loss": 4.550562858581543, "state_train_accuracy": 0.0, "executor_loss": 4.565143585205078, "loss": 37.49949264526367 }, { "run": "main_max24_curriculum_seed789", "arm": "max24_curriculum", "seed": 789, "stage": "stage1_max24", "stage_idx": 1, "local_step": 100, "global_step": 100, "compiler_max_steps": 24, "elapsed_sec": 278.332, "expansion_event": "init", "init_loss": 1.2008227109909058, "op_loss": 0.004859844222664833, "arg_loss": 0.5884870886802673, "state_loss": 2.3189282417297363, "state_train_accuracy": 0.8888888955116272, "executor_loss": 2.773819923400879, "loss": 12.254846572875977 }, { "run": "main_max24_curriculum_seed789", "arm": "max24_curriculum", "seed": 789, "stage": "stage1_max24", "stage_idx": 1, "local_step": 150, "global_step": 150, "compiler_max_steps": 24, "elapsed_sec": 416.772, "expansion_event": "init", "init_loss": 0.4457521438598633, "op_loss": 0.0001054686144925654, "arg_loss": 0.008442403748631477, "state_loss": 0.5287204384803772, "state_train_accuracy": 0.8484848737716675, "executor_loss": 0.47843730449676514, "loss": 2.8240413665771484 }, { "run": "main_max24_curriculum_seed789", "arm": "max24_curriculum", "seed": 789, "stage": "stage1_max24", "stage_idx": 1, "local_step": 200, "global_step": 200, "compiler_max_steps": 24, "elapsed_sec": 555.751, "expansion_event": "init", "init_loss": 0.023293089121580124, "op_loss": 8.538313522876706e-06, "arg_loss": 0.0008157340926118195, "state_loss": 0.018576906993985176, "state_train_accuracy": 1.0, "executor_loss": 0.025765426456928253, "loss": 0.14078617095947266 }, { "run": "main_max24_curriculum_seed789", "arm": "max24_curriculum", "seed": 789, "stage": "stage1_max24", "stage_idx": 1, "local_step": 250, "global_step": 250, "compiler_max_steps": 24, "elapsed_sec": 693.911, "expansion_event": "init", "init_loss": 0.009750641882419586, "op_loss": 3.397457248865976e-06, "arg_loss": 0.0001420224434696138, "state_loss": 0.005889313295483589, "state_train_accuracy": 1.0, "executor_loss": 0.01036825217306614, "loss": 0.05583161860704422 }, { "run": "main_max24_curriculum_seed789", "arm": "max24_curriculum", "seed": 789, "stage": "stage1_max24", "stage_idx": 1, "local_step": 300, "global_step": 300, "compiler_max_steps": 24, "elapsed_sec": 832.268, "expansion_event": "init", "init_loss": 0.0007331287488341331, "op_loss": 9.844903252087533e-05, "arg_loss": 0.0007424491923302412, "state_loss": 0.0035100053064525127, "state_train_accuracy": 1.0, "executor_loss": 0.005147892516106367, "loss": 0.01465865969657898 }, { "run": "main_max24_curriculum_seed789", "arm": "max24_curriculum", "seed": 789, "stage": "stage2_max24", "stage_idx": 2, "local_step": 1, "global_step": 301, "compiler_max_steps": 24, "elapsed_sec": 964.953, "expansion_event": "same", "init_loss": 0.8112020492553711, "op_loss": 4.265385359758511e-05, "arg_loss": 0.00048322355723939836, "state_loss": 1.1818569898605347, "state_train_accuracy": 0.8181818723678589, "executor_loss": 0.814486563205719, "loss": 5.243127346038818 }, { "run": "main_max24_curriculum_seed789", "arm": "max24_curriculum", "seed": 789, "stage": "stage2_max24", "stage_idx": 2, "local_step": 50, "global_step": 350, "compiler_max_steps": 24, "elapsed_sec": 1116.183, "expansion_event": "same", "init_loss": 0.0015412785578519106, "op_loss": 3.578553150873631e-05, "arg_loss": 0.0012224019737914205, "state_loss": 0.006334634963423014, "state_train_accuracy": 1.0, "executor_loss": 0.009247515350580215, "loss": 0.02667265757918358 }, { "run": "main_max24_curriculum_seed789", "arm": "max24_curriculum", "seed": 789, "stage": "stage2_max24", "stage_idx": 2, "local_step": 100, "global_step": 400, "compiler_max_steps": 24, "elapsed_sec": 1270.908, "expansion_event": "same", "init_loss": 0.0021139224991202354, "op_loss": 1.6882119098227122e-06, "arg_loss": 0.00020410858269315213, "state_loss": 0.0039866603910923, "state_train_accuracy": 1.0, "executor_loss": 0.0038632177747786045, "loss": 0.017123691737651825 }, { "run": "main_max24_curriculum_seed789", "arm": "max24_curriculum", "seed": 789, "stage": "stage2_max24", "stage_idx": 2, "local_step": 150, "global_step": 450, "compiler_max_steps": 24, "elapsed_sec": 1424.919, "expansion_event": "same", "init_loss": 0.0004696109681390226, "op_loss": 8.983265615825076e-07, "arg_loss": 4.05782884627115e-05, "state_loss": 0.000883891130797565, "state_train_accuracy": 1.0, "executor_loss": 0.0009049054351635277, "loss": 0.0038304519839584827 }, { "run": "main_max24_curriculum_seed789", "arm": "max24_curriculum", "seed": 789, "stage": "stage3_max24", "stage_idx": 3, "local_step": 1, "global_step": 451, "compiler_max_steps": 24, "elapsed_sec": 1558.394, "expansion_event": "same", "init_loss": 0.00020507023145910352, "op_loss": 0.08058735728263855, "arg_loss": 0.12780801951885223, "state_loss": 0.40770357847213745, "state_train_accuracy": 0.9640287756919861, "executor_loss": 1.3740800619125366, "loss": 2.3744232654571533 }, { "run": "main_max24_curriculum_seed789", "arm": "max24_curriculum", "seed": 789, "stage": "stage3_max24", "stage_idx": 3, "local_step": 50, "global_step": 500, "compiler_max_steps": 24, "elapsed_sec": 1726.034, "expansion_event": "same", "init_loss": 0.0007795548881404102, "op_loss": 2.6617547064233804e-06, "arg_loss": 3.979393659392372e-05, "state_loss": 0.0013686495367437601, "state_train_accuracy": 1.0, "executor_loss": 0.0015027717454358935, "loss": 0.006151478737592697 }, { "run": "main_max24_curriculum_seed789", "arm": "max24_curriculum", "seed": 789, "stage": "stage3_max24", "stage_idx": 3, "local_step": 100, "global_step": 550, "compiler_max_steps": 24, "elapsed_sec": 1898.796, "expansion_event": "same", "init_loss": 0.00013150523591320962, "op_loss": 2.897683771152515e-05, "arg_loss": 0.001162869157269597, "state_loss": 0.006931970827281475, "state_train_accuracy": 0.9999999403953552, "executor_loss": 0.017885275185108185, "loss": 0.030023720115423203 }, { "run": "main_max24_curriculum_seed789", "arm": "max24_curriculum", "seed": 789, "stage": "stage3_max24", "stage_idx": 3, "local_step": 150, "global_step": 600, "compiler_max_steps": 24, "elapsed_sec": 2070.469, "expansion_event": "same", "init_loss": 3.607469261623919e-05, "op_loss": 1.2759027413267177e-05, "arg_loss": 0.00013047915126662701, "state_loss": 0.0010484008817002177, "state_train_accuracy": 1.0, "executor_loss": 0.0017550359480082989, "loss": 0.0034824111498892307 }, { "run": "main_max24_curriculum_seed789", "arm": "max24_curriculum", "seed": 789, "stage": "stage3_max24", "stage_idx": 3, "local_step": 200, "global_step": 650, "compiler_max_steps": 24, "elapsed_sec": 2243.64, "expansion_event": "same", "init_loss": 0.00012247492850292474, "op_loss": 7.489198469556868e-05, "arg_loss": 0.00023563075228594244, "state_loss": 0.0023879935033619404, "state_train_accuracy": 1.0, "executor_loss": 0.005247139371931553, "loss": 0.009142447263002396 }, { "run": "main_max24_curriculum_seed789", "arm": "max24_curriculum", "seed": 789, "stage": "stage3_max24", "stage_idx": 3, "local_step": 250, "global_step": 700, "compiler_max_steps": 24, "elapsed_sec": 2416.337, "expansion_event": "same", "init_loss": 5.9883495850954205e-05, "op_loss": 2.952782551801647e-06, "arg_loss": 5.867933214176446e-05, "state_loss": 0.0006131496629677713, "state_train_accuracy": 1.0, "executor_loss": 0.0011307407403364778, "loss": 0.0022210944443941116 }, { "run": "main_max24_curriculum_seed789", "arm": "max24_curriculum", "seed": 789, "stage": "stage3_max24", "stage_idx": 3, "local_step": 300, "global_step": 750, "compiler_max_steps": 24, "elapsed_sec": 2589.026, "expansion_event": "same", "init_loss": 6.0242527979426086e-05, "op_loss": 7.308868589461781e-06, "arg_loss": 3.314349305583164e-05, "state_loss": 0.00035813014255836606, "state_train_accuracy": 0.9999999403953552, "executor_loss": 0.0006077036960050464, "loss": 0.0013466868549585342 } ], "metrics": [ { "stage": "stage1_max24", "split": "standard_L8", "n": 64, "max_steps": 24, "executor_accuracy": 0.984375, "executor_target_mass": 0.9780843928456306, "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_seed789", "arm": "max24_curriculum", "seed": 789, "global_step": 300, "stage_idx": 1 }, { "stage": "stage1_max24", "split": "paraphrase_L8", "n": 64, "max_steps": 24, "executor_accuracy": 0.953125, "executor_target_mass": 0.9467832371592522, "direct_accuracy": NaN, "init_accuracy": 0.953125, "op_accuracy": 1.0, "arg_accuracy": 1.0, "program_exact": 0.953125, "state_accuracy": 0.953125, "state_all_exact": 0.953125, "state_prefix_fraction": 0.953125, "state_mean_correct_prefix": 7.625, "run": "main_max24_curriculum_seed789", "arm": "max24_curriculum", "seed": 789, "global_step": 300, "stage_idx": 1 }, { "stage": "stage1_max24", "split": "heldout_L8", "n": 64, "max_steps": 24, "executor_accuracy": 0.75, "executor_target_mass": 0.7434692792594433, "direct_accuracy": NaN, "init_accuracy": 0.984375, "op_accuracy": 0.94921875, "arg_accuracy": 0.978515625, "program_exact": 0.75, "state_accuracy": 0.82421875, "state_all_exact": 0.75, "state_prefix_fraction": 0.82421875, "state_mean_correct_prefix": 6.59375, "run": "main_max24_curriculum_seed789", "arm": "max24_curriculum", "seed": 789, "global_step": 300, "stage_idx": 1 }, { "stage": "stage1_max24", "split": "paired_L8", "n": 64, "max_steps": 24, "executor_accuracy": 1.0, "executor_target_mass": 0.9902346432209015, "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_seed789", "arm": "max24_curriculum", "seed": 789, "global_step": 300, "stage_idx": 1 }, { "stage": "stage1_max24", "split": "paired_heldout_L8", "n": 64, "max_steps": 24, "executor_accuracy": 0.828125, "executor_target_mass": 0.8166379332542419, "direct_accuracy": NaN, "init_accuracy": 0.9375, "op_accuracy": 0.98046875, "arg_accuracy": 0.9921875, "program_exact": 0.8125, "state_accuracy": 0.880859375, "state_all_exact": 0.828125, "state_prefix_fraction": 0.880859375, "state_mean_correct_prefix": 7.046875, "executor_pair_answer_consistency": 0.78125, "executor_pair_both_correct": 0.71875, "compiler_pair_program_consistency": 0.75, "compiler_pair_state_consistency": 0.78125, "run": "main_max24_curriculum_seed789", "arm": "max24_curriculum", "seed": 789, "global_step": 300, "stage_idx": 1 }, { "stage": "stage1_max24", "split": "standard_L16", "n": 64, "max_steps": 24, "executor_accuracy": 0.546875, "executor_target_mass": 0.512266181409359, "direct_accuracy": NaN, "init_accuracy": 0.984375, "op_accuracy": 0.974609375, "arg_accuracy": 0.9482421875, "program_exact": 0.5, "state_accuracy": 0.900390625, "state_all_exact": 0.5, "state_prefix_fraction": 0.89453125, "state_mean_correct_prefix": 14.3125, "run": "main_max24_curriculum_seed789", "arm": "max24_curriculum", "seed": 789, "global_step": 300, "stage_idx": 1 }, { "stage": "stage1_max24", "split": "paraphrase_L16", "n": 64, "max_steps": 24, "executor_accuracy": 0.359375, "executor_target_mass": 0.3458257047459483, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.9501953125, "arg_accuracy": 0.921875, "program_exact": 0.34375, "state_accuracy": 0.8623046875, "state_all_exact": 0.34375, "state_prefix_fraction": 0.859375, "state_mean_correct_prefix": 13.75, "run": "main_max24_curriculum_seed789", "arm": "max24_curriculum", "seed": 789, "global_step": 300, "stage_idx": 1 }, { "stage": "stage1_max24", "split": "heldout_L16", "n": 64, "max_steps": 24, "executor_accuracy": 0.171875, "executor_target_mass": 0.15122895874083042, "direct_accuracy": NaN, "init_accuracy": 0.984375, "op_accuracy": 0.89453125, "arg_accuracy": 0.8564453125, "program_exact": 0.171875, "state_accuracy": 0.705078125, "state_all_exact": 0.171875, "state_prefix_fraction": 0.7041015625, "state_mean_correct_prefix": 11.265625, "run": "main_max24_curriculum_seed789", "arm": "max24_curriculum", "seed": 789, "global_step": 300, "stage_idx": 1 }, { "stage": "stage1_max24", "split": "paired_L16", "n": 64, "max_steps": 24, "executor_accuracy": 0.5, "executor_target_mass": 0.4837851282209158, "direct_accuracy": NaN, "init_accuracy": 0.9375, "op_accuracy": 0.9638671875, "arg_accuracy": 0.943359375, "program_exact": 0.484375, "state_accuracy": 0.8583984375, "state_all_exact": 0.484375, "state_prefix_fraction": 0.857421875, "state_mean_correct_prefix": 13.71875, "executor_pair_answer_consistency": 0.40625, "executor_pair_both_correct": 0.375, "compiler_pair_program_consistency": 0.375, "compiler_pair_state_consistency": 0.375, "run": "main_max24_curriculum_seed789", "arm": "max24_curriculum", "seed": 789, "global_step": 300, "stage_idx": 1 }, { "stage": "stage1_max24", "split": "paired_heldout_L16", "n": 64, "max_steps": 24, "executor_accuracy": 0.25, "executor_target_mass": 0.2524145011266228, "direct_accuracy": NaN, "init_accuracy": 0.90625, "op_accuracy": 0.912109375, "arg_accuracy": 0.8759765625, "program_exact": 0.25, "state_accuracy": 0.69140625, "state_all_exact": 0.25, "state_prefix_fraction": 0.6904296875, "state_mean_correct_prefix": 11.046875, "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_seed789", "arm": "max24_curriculum", "seed": 789, "global_step": 300, "stage_idx": 1 }, { "stage": "stage1_max24", "split": "standard_L24", "n": 64, "max_steps": 24, "executor_accuracy": 0.015625, "executor_target_mass": 0.02429692560690455, "direct_accuracy": NaN, "init_accuracy": 0.9375, "op_accuracy": 0.8626302083333334, "arg_accuracy": 0.7688802083333334, "program_exact": 0.015625, "state_accuracy": 0.60546875, "state_all_exact": 0.015625, "state_prefix_fraction": 0.59765625, "state_mean_correct_prefix": 14.34375, "run": "main_max24_curriculum_seed789", "arm": "max24_curriculum", "seed": 789, "global_step": 300, "stage_idx": 1 }, { "stage": "stage1_max24", "split": "paraphrase_L24", "n": 64, "max_steps": 24, "executor_accuracy": 0.0, "executor_target_mass": 0.008640865329653025, "direct_accuracy": NaN, "init_accuracy": 0.953125, "op_accuracy": 0.8157552083333334, "arg_accuracy": 0.7141927083333334, "program_exact": 0.0, "state_accuracy": 0.5911458333333334, "state_all_exact": 0.0, "state_prefix_fraction": 0.587890625, "state_mean_correct_prefix": 14.109375, "run": "main_max24_curriculum_seed789", "arm": "max24_curriculum", "seed": 789, "global_step": 300, "stage_idx": 1 }, { "stage": "stage1_max24", "split": "heldout_L24", "n": 64, "max_steps": 24, "executor_accuracy": 0.015625, "executor_target_mass": 0.009537733974866569, "direct_accuracy": NaN, "init_accuracy": 0.984375, "op_accuracy": 0.71484375, "arg_accuracy": 0.5774739583333334, "program_exact": 0.0, "state_accuracy": 0.4524739583333333, "state_all_exact": 0.0, "state_prefix_fraction": 0.44921875, "state_mean_correct_prefix": 10.78125, "run": "main_max24_curriculum_seed789", "arm": "max24_curriculum", "seed": 789, "global_step": 300, "stage_idx": 1 }, { "stage": "stage1_max24", "split": "paired_L24", "n": 64, "max_steps": 24, "executor_accuracy": 0.015625, "executor_target_mass": 0.022693348379107192, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.8235677083333334, "arg_accuracy": 0.7174479166666666, "program_exact": 0.015625, "state_accuracy": 0.6139322916666666, "state_all_exact": 0.015625, "state_prefix_fraction": 0.6087239583333334, "state_mean_correct_prefix": 14.609375, "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_seed789", "arm": "max24_curriculum", "seed": 789, "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": 0.02316651912406087, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.7942708333333334, "arg_accuracy": 0.685546875, "program_exact": 0.015625, "state_accuracy": 0.572265625, "state_all_exact": 0.015625, "state_prefix_fraction": 0.5709635416666666, "state_mean_correct_prefix": 13.703125, "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_seed789", "arm": "max24_curriculum", "seed": 789, "global_step": 300, "stage_idx": 1 }, { "stage": "stage2_max24", "split": "standard_L8", "n": 64, "max_steps": 24, "executor_accuracy": 1.0, "executor_target_mass": 0.9993028119206429, "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_seed789", "arm": "max24_curriculum", "seed": 789, "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max24", "split": "paraphrase_L8", "n": 64, "max_steps": 24, "executor_accuracy": 1.0, "executor_target_mass": 0.9993355125188828, "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_seed789", "arm": "max24_curriculum", "seed": 789, "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max24", "split": "heldout_L8", "n": 64, "max_steps": 24, "executor_accuracy": 0.953125, "executor_target_mass": 0.9523302540183067, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.990234375, "arg_accuracy": 0.984375, "program_exact": 0.953125, "state_accuracy": 0.974609375, "state_all_exact": 0.953125, "state_prefix_fraction": 0.974609375, "state_mean_correct_prefix": 7.796875, "run": "main_max24_curriculum_seed789", "arm": "max24_curriculum", "seed": 789, "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max24", "split": "paired_L8", "n": 64, "max_steps": 24, "executor_accuracy": 1.0, "executor_target_mass": 0.9992759227752686, "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_seed789", "arm": "max24_curriculum", "seed": 789, "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max24", "split": "paired_heldout_L8", "n": 64, "max_steps": 24, "executor_accuracy": 0.984375, "executor_target_mass": 0.9833489954471588, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.994140625, "arg_accuracy": 0.994140625, "program_exact": 0.984375, "state_accuracy": 0.9921875, "state_all_exact": 0.984375, "state_prefix_fraction": 0.9921875, "state_mean_correct_prefix": 7.9375, "executor_pair_answer_consistency": 0.96875, "executor_pair_both_correct": 0.96875, "compiler_pair_program_consistency": 0.96875, "compiler_pair_state_consistency": 0.96875, "run": "main_max24_curriculum_seed789", "arm": "max24_curriculum", "seed": 789, "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max24", "split": "standard_L16", "n": 64, "max_steps": 24, "executor_accuracy": 1.0, "executor_target_mass": 0.9990491420030594, "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_seed789", "arm": "max24_curriculum", "seed": 789, "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max24", "split": "paraphrase_L16", "n": 64, "max_steps": 24, "executor_accuracy": 1.0, "executor_target_mass": 0.9988981857895851, "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_seed789", "arm": "max24_curriculum", "seed": 789, "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max24", "split": "heldout_L16", "n": 64, "max_steps": 24, "executor_accuracy": 0.96875, "executor_target_mass": 0.962573416531086, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 1.0, "arg_accuracy": 0.998046875, "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_seed789", "arm": "max24_curriculum", "seed": 789, "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max24", "split": "paired_L16", "n": 64, "max_steps": 24, "executor_accuracy": 1.0, "executor_target_mass": 0.9989942014217377, "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_seed789", "arm": "max24_curriculum", "seed": 789, "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max24", "split": "paired_heldout_L16", "n": 64, "max_steps": 24, "executor_accuracy": 0.96875, "executor_target_mass": 0.9604249596595764, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.9814453125, "arg_accuracy": 0.9755859375, "program_exact": 0.96875, "state_accuracy": 0.9755859375, "state_all_exact": 0.96875, "state_prefix_fraction": 0.9755859375, "state_mean_correct_prefix": 15.609375, "executor_pair_answer_consistency": 0.9375, "executor_pair_both_correct": 0.9375, "compiler_pair_program_consistency": 0.9375, "compiler_pair_state_consistency": 0.9375, "run": "main_max24_curriculum_seed789", "arm": "max24_curriculum", "seed": 789, "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max24", "split": "standard_L24", "n": 64, "max_steps": 24, "executor_accuracy": 0.578125, "executor_target_mass": 0.5634459126740694, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.9895833333333334, "arg_accuracy": 0.9772135416666666, "program_exact": 0.578125, "state_accuracy": 0.9446614583333334, "state_all_exact": 0.578125, "state_prefix_fraction": 0.9446614583333334, "state_mean_correct_prefix": 22.671875, "run": "main_max24_curriculum_seed789", "arm": "max24_curriculum", "seed": 789, "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max24", "split": "paraphrase_L24", "n": 64, "max_steps": 24, "executor_accuracy": 0.4375, "executor_target_mass": 0.43842961080372334, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.9739583333333334, "arg_accuracy": 0.955078125, "program_exact": 0.4375, "state_accuracy": 0.9055989583333334, "state_all_exact": 0.4375, "state_prefix_fraction": 0.9055989583333334, "state_mean_correct_prefix": 21.734375, "run": "main_max24_curriculum_seed789", "arm": "max24_curriculum", "seed": 789, "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max24", "split": "heldout_L24", "n": 64, "max_steps": 24, "executor_accuracy": 0.234375, "executor_target_mass": 0.23407111274855197, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.939453125, "arg_accuracy": 0.904296875, "program_exact": 0.234375, "state_accuracy": 0.8235677083333334, "state_all_exact": 0.234375, "state_prefix_fraction": 0.8235677083333334, "state_mean_correct_prefix": 19.765625, "run": "main_max24_curriculum_seed789", "arm": "max24_curriculum", "seed": 789, "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max24", "split": "paired_L24", "n": 64, "max_steps": 24, "executor_accuracy": 0.515625, "executor_target_mass": 0.5103201977908611, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.982421875, "arg_accuracy": 0.9641927083333334, "program_exact": 0.515625, "state_accuracy": 0.9375, "state_all_exact": 0.515625, "state_prefix_fraction": 0.9375, "state_mean_correct_prefix": 22.5, "executor_pair_answer_consistency": 0.15625, "executor_pair_both_correct": 0.15625, "compiler_pair_program_consistency": 0.15625, "compiler_pair_state_consistency": 0.15625, "run": "main_max24_curriculum_seed789", "arm": "max24_curriculum", "seed": 789, "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max24", "split": "paired_heldout_L24", "n": 64, "max_steps": 24, "executor_accuracy": 0.375, "executor_target_mass": 0.3616152200847864, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.9615885416666666, "arg_accuracy": 0.93359375, "program_exact": 0.359375, "state_accuracy": 0.8645833333333334, "state_all_exact": 0.359375, "state_prefix_fraction": 0.8639322916666666, "state_mean_correct_prefix": 20.734375, "executor_pair_answer_consistency": 0.09375, "executor_pair_both_correct": 0.09375, "compiler_pair_program_consistency": 0.09375, "compiler_pair_state_consistency": 0.09375, "run": "main_max24_curriculum_seed789", "arm": "max24_curriculum", "seed": 789, "global_step": 450, "stage_idx": 2 }, { "stage": "stage3_max24", "split": "standard_L8", "n": 64, "max_steps": 24, "executor_accuracy": 1.0, "executor_target_mass": 0.9996054694056511, "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_seed789", "arm": "max24_curriculum", "seed": 789, "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "paraphrase_L8", "n": 64, "max_steps": 24, "executor_accuracy": 1.0, "executor_target_mass": 0.9996984973549843, "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_seed789", "arm": "max24_curriculum", "seed": 789, "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "heldout_L8", "n": 64, "max_steps": 24, "executor_accuracy": 1.0, "executor_target_mass": 0.9993459805846214, "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_seed789", "arm": "max24_curriculum", "seed": 789, "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "paired_L8", "n": 64, "max_steps": 24, "executor_accuracy": 1.0, "executor_target_mass": 0.9996797516942024, "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_seed789", "arm": "max24_curriculum", "seed": 789, "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": 0.9996741786599159, "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_seed789", "arm": "max24_curriculum", "seed": 789, "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "standard_L16", "n": 64, "max_steps": 24, "executor_accuracy": 1.0, "executor_target_mass": 0.9993549287319183, "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_seed789", "arm": "max24_curriculum", "seed": 789, "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "paraphrase_L16", "n": 64, "max_steps": 24, "executor_accuracy": 1.0, "executor_target_mass": 0.9993212297558784, "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_seed789", "arm": "max24_curriculum", "seed": 789, "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "heldout_L16", "n": 64, "max_steps": 24, "executor_accuracy": 1.0, "executor_target_mass": 0.99812401086092, "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_seed789", "arm": "max24_curriculum", "seed": 789, "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "paired_L16", "n": 64, "max_steps": 24, "executor_accuracy": 1.0, "executor_target_mass": 0.9993072152137756, "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_seed789", "arm": "max24_curriculum", "seed": 789, "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "paired_heldout_L16", "n": 64, "max_steps": 24, "executor_accuracy": 1.0, "executor_target_mass": 0.9985946789383888, "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_seed789", "arm": "max24_curriculum", "seed": 789, "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "standard_L24", "n": 64, "max_steps": 24, "executor_accuracy": 0.171875, "executor_target_mass": 0.169042355613783, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.9186197916666666, "arg_accuracy": 0.89453125, "program_exact": 0.140625, "state_accuracy": 0.853515625, "state_all_exact": 0.140625, "state_prefix_fraction": 0.845703125, "state_mean_correct_prefix": 20.296875, "run": "main_max24_curriculum_seed789", "arm": "max24_curriculum", "seed": 789, "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "paraphrase_L24", "n": 64, "max_steps": 24, "executor_accuracy": 0.984375, "executor_target_mass": 0.9824599698185921, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.9993489583333334, "arg_accuracy": 0.9993489583333334, "program_exact": 0.984375, "state_accuracy": 0.9993489583333334, "state_all_exact": 0.984375, "state_prefix_fraction": 0.9993489583333334, "state_mean_correct_prefix": 23.984375, "run": "main_max24_curriculum_seed789", "arm": "max24_curriculum", "seed": 789, "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "heldout_L24", "n": 64, "max_steps": 24, "executor_accuracy": 0.921875, "executor_target_mass": 0.9104860201478004, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.9973958333333334, "arg_accuracy": 0.9928385416666666, "program_exact": 0.921875, "state_accuracy": 0.9869791666666666, "state_all_exact": 0.921875, "state_prefix_fraction": 0.9869791666666666, "state_mean_correct_prefix": 23.6875, "run": "main_max24_curriculum_seed789", "arm": "max24_curriculum", "seed": 789, "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "paired_L24", "n": 64, "max_steps": 24, "executor_accuracy": 0.609375, "executor_target_mass": 0.6032218299806118, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.9713541666666666, "arg_accuracy": 0.9524739583333334, "program_exact": 0.59375, "state_accuracy": 0.9322916666666666, "state_all_exact": 0.59375, "state_prefix_fraction": 0.927734375, "state_mean_correct_prefix": 22.265625, "executor_pair_answer_consistency": 0.21875, "executor_pair_both_correct": 0.21875, "compiler_pair_program_consistency": 0.1875, "compiler_pair_state_consistency": 0.1875, "run": "main_max24_curriculum_seed789", "arm": "max24_curriculum", "seed": 789, "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "paired_heldout_L24", "n": 64, "max_steps": 24, "executor_accuracy": 0.46875, "executor_target_mass": 0.46442194655537605, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.9453125, "arg_accuracy": 0.9290364583333334, "program_exact": 0.453125, "state_accuracy": 0.9016927083333334, "state_all_exact": 0.453125, "state_prefix_fraction": 0.8990885416666666, "state_mean_correct_prefix": 21.578125, "executor_pair_answer_consistency": 0.25, "executor_pair_both_correct": 0.21875, "compiler_pair_program_consistency": 0.21875, "compiler_pair_state_consistency": 0.21875, "run": "main_max24_curriculum_seed789", "arm": "max24_curriculum", "seed": 789, "global_step": 750, "stage_idx": 3 } ], "checkpoints": [ { "run": "main_max24_curriculum_seed789", "seed": 789, "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_seed789/stage1_max24_step00300", "model_checkpoint": "/workspace/large_artifacts/qwen_compiler_multiseed_reattribution/checkpoints/main_max24_curriculum_seed789/stage1_max24_step00300/adapter", "heads_checkpoint": "/workspace/large_artifacts/qwen_compiler_multiseed_reattribution/checkpoints/main_max24_curriculum_seed789/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": 0.015625, "state_prefix_fraction": 0.6087239583333334 }, { "run": "main_max24_curriculum_seed789", "seed": 789, "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_seed789/stage2_max24_step00450", "model_checkpoint": "/workspace/large_artifacts/qwen_compiler_multiseed_reattribution/checkpoints/main_max24_curriculum_seed789/stage2_max24_step00450/adapter", "heads_checkpoint": "/workspace/large_artifacts/qwen_compiler_multiseed_reattribution/checkpoints/main_max24_curriculum_seed789/stage2_max24_step00450/heads.pt", "selection_split": "paired_L24", "selection_metric": "executor_pair_both_correct", "selection_value": 0.15625, "executor_accuracy": 0.515625, "program_exact": 0.515625, "state_prefix_fraction": 0.9375 }, { "run": "main_max24_curriculum_seed789", "seed": 789, "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_seed789/stage3_max24_step00750", "model_checkpoint": "/workspace/large_artifacts/qwen_compiler_multiseed_reattribution/checkpoints/main_max24_curriculum_seed789/stage3_max24_step00750/adapter", "heads_checkpoint": "/workspace/large_artifacts/qwen_compiler_multiseed_reattribution/checkpoints/main_max24_curriculum_seed789/stage3_max24_step00750/heads.pt", "selection_split": "paired_L24", "selection_metric": "executor_pair_both_correct", "selection_value": 0.21875, "executor_accuracy": 0.609375, "program_exact": 0.59375, "state_prefix_fraction": 0.927734375 } ], "elapsed_sec": 2716.749 }