{ "run": "main_expand_copy_seed456", "arm": "expand_copy", "seed": 456, "args": { "run_name": "main_expand_copy_seed456", "arm": "expand_copy", "model_id": "Qwen/Qwen3-4B", "seed": 456, "modulus": 97, "stage_max_steps": "8,16,24", "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_max8", "max_steps": 8, "train_min_len": 1, "train_max_len": 8, "steps": 300 }, { "name": "stage2_max16", "max_steps": 16, "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_expand_copy_seed456", "arm": "expand_copy", "seed": 456, "stage": "stage1_max8", "stage_idx": 1, "local_step": 1, "global_step": 1, "compiler_max_steps": 8, "elapsed_sec": 2.577, "expansion_event": "init", "init_loss": 4.445868015289307, "op_loss": 1.104657769203186, "arg_loss": 4.564539432525635, "state_loss": 4.577871322631836, "state_train_accuracy": 0.0, "executor_loss": 4.578193187713623, "loss": 46.30235290527344 }, { "run": "main_expand_copy_seed456", "arm": "expand_copy", "seed": 456, "stage": "stage1_max8", "stage_idx": 1, "local_step": 50, "global_step": 50, "compiler_max_steps": 8, "elapsed_sec": 56.805, "expansion_event": "init", "init_loss": 3.860687732696533, "op_loss": 1.1128703355789185, "arg_loss": 3.5231008529663086, "state_loss": 4.583004951477051, "state_train_accuracy": 0.0, "executor_loss": 4.577595233917236, "loss": 39.808624267578125 }, { "run": "main_expand_copy_seed456", "arm": "expand_copy", "seed": 456, "stage": "stage1_max8", "stage_idx": 1, "local_step": 100, "global_step": 100, "compiler_max_steps": 8, "elapsed_sec": 112.553, "expansion_event": "init", "init_loss": 1.0481586456298828, "op_loss": 0.3059787452220917, "arg_loss": 2.0089938640594482, "state_loss": 3.9879884719848633, "state_train_accuracy": 0.25, "executor_loss": 4.340097427368164, "loss": 20.862674713134766 }, { "run": "main_expand_copy_seed456", "arm": "expand_copy", "seed": 456, "stage": "stage1_max8", "stage_idx": 1, "local_step": 150, "global_step": 150, "compiler_max_steps": 8, "elapsed_sec": 168.632, "expansion_event": "init", "init_loss": 0.22482845187187195, "op_loss": 0.00039889541221782565, "arg_loss": 0.03283090889453888, "state_loss": 0.3483756482601166, "state_train_accuracy": 1.0, "executor_loss": 0.41447821259498596, "loss": 1.7938902378082275 }, { "run": "main_expand_copy_seed456", "arm": "expand_copy", "seed": 456, "stage": "stage1_max8", "stage_idx": 1, "local_step": 200, "global_step": 200, "compiler_max_steps": 8, "elapsed_sec": 224.907, "expansion_event": "init", "init_loss": 0.027010779827833176, "op_loss": 3.7781828723382205e-05, "arg_loss": 0.0038600999396294355, "state_loss": 0.040959056466817856, "state_train_accuracy": 1.0, "executor_loss": 0.04453548043966293, "loss": 0.2090158462524414 }, { "run": "main_expand_copy_seed456", "arm": "expand_copy", "seed": 456, "stage": "stage1_max8", "stage_idx": 1, "local_step": 250, "global_step": 250, "compiler_max_steps": 8, "elapsed_sec": 280.741, "expansion_event": "init", "init_loss": 0.005293854512274265, "op_loss": 3.9221104088937864e-05, "arg_loss": 0.0006787513266317546, "state_loss": 0.007327323779463768, "state_train_accuracy": 1.0, "executor_loss": 0.008516349829733372, "loss": 0.03977331519126892 }, { "run": "main_expand_copy_seed456", "arm": "expand_copy", "seed": 456, "stage": "stage1_max8", "stage_idx": 1, "local_step": 300, "global_step": 300, "compiler_max_steps": 8, "elapsed_sec": 336.524, "expansion_event": "init", "init_loss": 0.0008064471185207367, "op_loss": 2.356366849198821e-06, "arg_loss": 0.00016434400458820164, "state_loss": 0.0013235642109066248, "state_train_accuracy": 1.0, "executor_loss": 0.0014315752778202295, "loss": 0.00664066057652235 }, { "run": "main_expand_copy_seed456", "arm": "expand_copy", "seed": 456, "stage": "stage2_max16", "stage_idx": 2, "local_step": 1, "global_step": 301, "compiler_max_steps": 16, "elapsed_sec": 356.972, "expansion_event": "expand_8_to_16_copy_last", "init_loss": 0.08159530162811279, "op_loss": 0.7577157020568848, "arg_loss": 0.9061954021453857, "state_loss": 1.6203571557998657, "state_train_accuracy": 0.8289473652839661, "executor_loss": 3.4164576530456543, "loss": 9.74569320678711 }, { "run": "main_expand_copy_seed456", "arm": "expand_copy", "seed": 456, "stage": "stage2_max16", "stage_idx": 2, "local_step": 50, "global_step": 350, "compiler_max_steps": 16, "elapsed_sec": 465.875, "expansion_event": "expand_8_to_16_copy_last", "init_loss": 0.0009365326841361821, "op_loss": 3.618208666011924e-06, "arg_loss": 0.00014506485604215413, "state_loss": 0.0019581655506044626, "state_train_accuracy": 1.0, "executor_loss": 0.002932559233158827, "loss": 0.009220733307301998 }, { "run": "main_expand_copy_seed456", "arm": "expand_copy", "seed": 456, "stage": "stage2_max16", "stage_idx": 2, "local_step": 100, "global_step": 400, "compiler_max_steps": 16, "elapsed_sec": 576.1, "expansion_event": "expand_8_to_16_copy_last", "init_loss": 0.00010016023588832468, "op_loss": 4.454661564068374e-07, "arg_loss": 2.6266394343110733e-05, "state_loss": 0.00021009727788623422, "state_train_accuracy": 1.0, "executor_loss": 0.000227080105105415, "loss": 0.0009433294180780649 }, { "run": "main_expand_copy_seed456", "arm": "expand_copy", "seed": 456, "stage": "stage2_max16", "stage_idx": 2, "local_step": 150, "global_step": 450, "compiler_max_steps": 16, "elapsed_sec": 686.679, "expansion_event": "expand_8_to_16_copy_last", "init_loss": 9.022174344863743e-05, "op_loss": 2.4951836167019792e-05, "arg_loss": 0.0001602686825208366, "state_loss": 0.0008643088513053954, "state_train_accuracy": 1.0, "executor_loss": 0.0016415114514529705, "loss": 0.003532733768224716 }, { "run": "main_expand_copy_seed456", "arm": "expand_copy", "seed": 456, "stage": "stage3_max24", "stage_idx": 3, "local_step": 1, "global_step": 451, "compiler_max_steps": 24, "elapsed_sec": 750.186, "expansion_event": "expand_16_to_24_copy_last", "init_loss": 0.0004770579398609698, "op_loss": 0.2800924777984619, "arg_loss": 0.5748658180236816, "state_loss": 0.6099792718887329, "state_train_accuracy": 0.9338842630386353, "executor_loss": 2.240719795227051, "loss": 5.432162761688232 }, { "run": "main_expand_copy_seed456", "arm": "expand_copy", "seed": 456, "stage": "stage3_max24", "stage_idx": 3, "local_step": 50, "global_step": 500, "compiler_max_steps": 24, "elapsed_sec": 920.176, "expansion_event": "expand_16_to_24_copy_last", "init_loss": 7.575334166176617e-05, "op_loss": 0.1738293468952179, "arg_loss": 0.06120062991976738, "state_loss": 0.16692781448364258, "state_train_accuracy": 0.9790209531784058, "executor_loss": 2.656620502471924, "loss": 3.242483139038086 }, { "run": "main_expand_copy_seed456", "arm": "expand_copy", "seed": 456, "stage": "stage3_max24", "stage_idx": 3, "local_step": 100, "global_step": 550, "compiler_max_steps": 24, "elapsed_sec": 1092.695, "expansion_event": "expand_16_to_24_copy_last", "init_loss": 0.0011062058620154858, "op_loss": 0.8129562735557556, "arg_loss": 1.340968370437622, "state_loss": 1.3341846466064453, "state_train_accuracy": 0.8181818127632141, "executor_loss": 2.3263232707977295, "loss": 9.84176254272461 }, { "run": "main_expand_copy_seed456", "arm": "expand_copy", "seed": 456, "stage": "stage3_max24", "stage_idx": 3, "local_step": 150, "global_step": 600, "compiler_max_steps": 24, "elapsed_sec": 1265.546, "expansion_event": "expand_16_to_24_copy_last", "init_loss": 4.4016705942340195e-05, "op_loss": 7.85201191320084e-05, "arg_loss": 0.024189764633774757, "state_loss": 0.04979856312274933, "state_train_accuracy": 0.9819819927215576, "executor_loss": 0.33642327785491943, "loss": 0.4832354784011841 }, { "run": "main_expand_copy_seed456", "arm": "expand_copy", "seed": 456, "stage": "stage3_max24", "stage_idx": 3, "local_step": 200, "global_step": 650, "compiler_max_steps": 24, "elapsed_sec": 1438.991, "expansion_event": "expand_16_to_24_copy_last", "init_loss": 0.0001348788500763476, "op_loss": 2.0318937458796427e-05, "arg_loss": 0.0001210309419548139, "state_loss": 0.0012850988423451781, "state_train_accuracy": 0.9999999403953552, "executor_loss": 0.00230239680968225, "loss": 0.004631453659385443 }, { "run": "main_expand_copy_seed456", "arm": "expand_copy", "seed": 456, "stage": "stage3_max24", "stage_idx": 3, "local_step": 250, "global_step": 700, "compiler_max_steps": 24, "elapsed_sec": 1612.928, "expansion_event": "expand_16_to_24_copy_last", "init_loss": 2.357329140068032e-05, "op_loss": 1.8561361230240436e-06, "arg_loss": 2.825941555784084e-05, "state_loss": 0.0003225062391720712, "state_train_accuracy": 1.0, "executor_loss": 0.0005582475569099188, "loss": 0.0010899407789111137 }, { "run": "main_expand_copy_seed456", "arm": "expand_copy", "seed": 456, "stage": "stage3_max24", "stage_idx": 3, "local_step": 300, "global_step": 750, "compiler_max_steps": 24, "elapsed_sec": 1786.939, "expansion_event": "expand_16_to_24_copy_last", "init_loss": 2.1486834157258272e-05, "op_loss": 2.2080707822169643e-06, "arg_loss": 8.846446871757507e-05, "state_loss": 0.00028116710018366575, "state_train_accuracy": 1.0, "executor_loss": 0.0014496208168566227, "loss": 0.0021728011779487133 } ], "metrics": [ { "stage": "stage1_max8", "split": "standard_L8", "n": 64, "max_steps": 8, "executor_accuracy": 0.984375, "executor_target_mass": 0.9654683321714401, "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_expand_copy_seed456", "arm": "expand_copy", "seed": 456, "global_step": 300, "stage_idx": 1 }, { "stage": "stage1_max8", "split": "paraphrase_L8", "n": 64, "max_steps": 8, "executor_accuracy": 1.0, "executor_target_mass": 0.9824669733643532, "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_expand_copy_seed456", "arm": "expand_copy", "seed": 456, "global_step": 300, "stage_idx": 1 }, { "stage": "stage1_max8", "split": "heldout_L8", "n": 64, "max_steps": 8, "executor_accuracy": 0.734375, "executor_target_mass": 0.7271504253149033, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.919921875, "arg_accuracy": 0.88671875, "program_exact": 0.71875, "state_accuracy": 0.845703125, "state_all_exact": 0.71875, "state_prefix_fraction": 0.84375, "state_mean_correct_prefix": 6.75, "run": "main_expand_copy_seed456", "arm": "expand_copy", "seed": 456, "global_step": 300, "stage_idx": 1 }, { "stage": "stage1_max8", "split": "paired_L8", "n": 64, "max_steps": 8, "executor_accuracy": 1.0, "executor_target_mass": 0.9923444762825966, "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_expand_copy_seed456", "arm": "expand_copy", "seed": 456, "global_step": 300, "stage_idx": 1 }, { "stage": "stage1_max8", "split": "paired_heldout_L8", "n": 64, "max_steps": 8, "executor_accuracy": 0.859375, "executor_target_mass": 0.862051248550415, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.966796875, "arg_accuracy": 0.94140625, "program_exact": 0.859375, "state_accuracy": 0.921875, "state_all_exact": 0.859375, "state_prefix_fraction": 0.921875, "state_mean_correct_prefix": 7.375, "executor_pair_answer_consistency": 0.71875, "executor_pair_both_correct": 0.71875, "compiler_pair_program_consistency": 0.71875, "compiler_pair_state_consistency": 0.71875, "run": "main_expand_copy_seed456", "arm": "expand_copy", "seed": 456, "global_step": 300, "stage_idx": 1 }, { "stage": "stage2_max16", "split": "standard_L8", "n": 64, "max_steps": 16, "executor_accuracy": 1.0, "executor_target_mass": 0.9941795095801353, "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_expand_copy_seed456", "arm": "expand_copy", "seed": 456, "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max16", "split": "paraphrase_L8", "n": 64, "max_steps": 16, "executor_accuracy": 1.0, "executor_target_mass": 0.9989743158221245, "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_expand_copy_seed456", "arm": "expand_copy", "seed": 456, "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max16", "split": "heldout_L8", "n": 64, "max_steps": 16, "executor_accuracy": 0.9375, "executor_target_mass": 0.9130064845085144, "direct_accuracy": NaN, "init_accuracy": 0.953125, "op_accuracy": 0.99609375, "arg_accuracy": 1.0, "program_exact": 0.9375, "state_accuracy": 0.939453125, "state_all_exact": 0.9375, "state_prefix_fraction": 0.939453125, "state_mean_correct_prefix": 7.515625, "run": "main_expand_copy_seed456", "arm": "expand_copy", "seed": 456, "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max16", "split": "paired_L8", "n": 64, "max_steps": 16, "executor_accuracy": 1.0, "executor_target_mass": 0.999245710670948, "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_expand_copy_seed456", "arm": "expand_copy", "seed": 456, "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max16", "split": "paired_heldout_L8", "n": 64, "max_steps": 16, "executor_accuracy": 0.921875, "executor_target_mass": 0.9232829287648201, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.9921875, "arg_accuracy": 0.986328125, "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_expand_copy_seed456", "arm": "expand_copy", "seed": 456, "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max16", "split": "standard_L16", "n": 64, "max_steps": 16, "executor_accuracy": 1.0, "executor_target_mass": 0.9917814508080482, "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_expand_copy_seed456", "arm": "expand_copy", "seed": 456, "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max16", "split": "paraphrase_L16", "n": 64, "max_steps": 16, "executor_accuracy": 0.984375, "executor_target_mass": 0.9833333119750023, "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": 15.75, "run": "main_expand_copy_seed456", "arm": "expand_copy", "seed": 456, "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max16", "split": "heldout_L16", "n": 64, "max_steps": 16, "executor_accuracy": 0.875, "executor_target_mass": 0.7792706936597824, "direct_accuracy": NaN, "init_accuracy": 0.984375, "op_accuracy": 0.9814453125, "arg_accuracy": 0.9853515625, "program_exact": 0.875, "state_accuracy": 0.9345703125, "state_all_exact": 0.875, "state_prefix_fraction": 0.9345703125, "state_mean_correct_prefix": 14.953125, "run": "main_expand_copy_seed456", "arm": "expand_copy", "seed": 456, "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max16", "split": "paired_L16", "n": 64, "max_steps": 16, "executor_accuracy": 1.0, "executor_target_mass": 0.9978031143546104, "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_expand_copy_seed456", "arm": "expand_copy", "seed": 456, "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max16", "split": "paired_heldout_L16", "n": 64, "max_steps": 16, "executor_accuracy": 0.921875, "executor_target_mass": 0.8882795125246048, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.990234375, "arg_accuracy": 0.994140625, "program_exact": 0.921875, "state_accuracy": 0.98046875, "state_all_exact": 0.921875, "state_prefix_fraction": 0.98046875, "state_mean_correct_prefix": 15.6875, "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_expand_copy_seed456", "arm": "expand_copy", "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": 0.9998933672904968, "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_expand_copy_seed456", "arm": "expand_copy", "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": 0.999892346560955, "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_expand_copy_seed456", "arm": "expand_copy", "seed": 456, "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "heldout_L8", "n": 64, "max_steps": 24, "executor_accuracy": 0.90625, "executor_target_mass": 0.9010856077075005, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.984375, "arg_accuracy": 0.9921875, "program_exact": 0.90625, "state_accuracy": 0.939453125, "state_all_exact": 0.90625, "state_prefix_fraction": 0.939453125, "state_mean_correct_prefix": 7.515625, "run": "main_expand_copy_seed456", "arm": "expand_copy", "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": 0.9998984858393669, "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_expand_copy_seed456", "arm": "expand_copy", "seed": 456, "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "paired_heldout_L8", "n": 64, "max_steps": 24, "executor_accuracy": 0.96875, "executor_target_mass": 0.9620997086167336, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.994140625, "arg_accuracy": 0.994140625, "program_exact": 0.96875, "state_accuracy": 0.986328125, "state_all_exact": 0.96875, "state_prefix_fraction": 0.986328125, "state_mean_correct_prefix": 7.890625, "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_expand_copy_seed456", "arm": "expand_copy", "seed": 456, "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "standard_L16", "n": 64, "max_steps": 24, "executor_accuracy": 1.0, "executor_target_mass": 0.9997889995574951, "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_expand_copy_seed456", "arm": "expand_copy", "seed": 456, "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "paraphrase_L16", "n": 64, "max_steps": 24, "executor_accuracy": 1.0, "executor_target_mass": 0.9997904971241951, "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_expand_copy_seed456", "arm": "expand_copy", "seed": 456, "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "heldout_L16", "n": 64, "max_steps": 24, "executor_accuracy": 0.828125, "executor_target_mass": 0.831321232020855, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.9716796875, "arg_accuracy": 0.951171875, "program_exact": 0.828125, "state_accuracy": 0.908203125, "state_all_exact": 0.828125, "state_prefix_fraction": 0.908203125, "state_mean_correct_prefix": 14.53125, "run": "main_expand_copy_seed456", "arm": "expand_copy", "seed": 456, "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "paired_L16", "n": 64, "max_steps": 24, "executor_accuracy": 1.0, "executor_target_mass": 0.9997943043708801, "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_expand_copy_seed456", "arm": "expand_copy", "seed": 456, "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "paired_heldout_L16", "n": 64, "max_steps": 24, "executor_accuracy": 0.890625, "executor_target_mass": 0.8924631029367447, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.9619140625, "arg_accuracy": 0.95703125, "program_exact": 0.890625, "state_accuracy": 0.9228515625, "state_all_exact": 0.890625, "state_prefix_fraction": 0.9228515625, "state_mean_correct_prefix": 14.765625, "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_expand_copy_seed456", "arm": "expand_copy", "seed": 456, "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "standard_L24", "n": 64, "max_steps": 24, "executor_accuracy": 0.484375, "executor_target_mass": 0.4858877956867218, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.9694010416666666, "arg_accuracy": 0.95703125, "program_exact": 0.453125, "state_accuracy": 0.9524739583333334, "state_all_exact": 0.453125, "state_prefix_fraction": 0.951171875, "state_mean_correct_prefix": 22.828125, "run": "main_expand_copy_seed456", "arm": "expand_copy", "seed": 456, "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "paraphrase_L24", "n": 64, "max_steps": 24, "executor_accuracy": 0.734375, "executor_target_mass": 0.7317011952400208, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.982421875, "arg_accuracy": 0.974609375, "program_exact": 0.71875, "state_accuracy": 0.9694010416666666, "state_all_exact": 0.71875, "state_prefix_fraction": 0.9680989583333334, "state_mean_correct_prefix": 23.234375, "run": "main_expand_copy_seed456", "arm": "expand_copy", "seed": 456, "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "heldout_L24", "n": 64, "max_steps": 24, "executor_accuracy": 0.078125, "executor_target_mass": 0.07719126716256142, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.8717447916666666, "arg_accuracy": 0.8235677083333334, "program_exact": 0.078125, "state_accuracy": 0.751953125, "state_all_exact": 0.078125, "state_prefix_fraction": 0.7493489583333334, "state_mean_correct_prefix": 17.984375, "run": "main_expand_copy_seed456", "arm": "expand_copy", "seed": 456, "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "paired_L24", "n": 64, "max_steps": 24, "executor_accuracy": 0.546875, "executor_target_mass": 0.5534789264202118, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.9765625, "arg_accuracy": 0.9609375, "program_exact": 0.546875, "state_accuracy": 0.953125, "state_all_exact": 0.546875, "state_prefix_fraction": 0.953125, "state_mean_correct_prefix": 22.875, "executor_pair_answer_consistency": 0.21875, "executor_pair_both_correct": 0.21875, "compiler_pair_program_consistency": 0.21875, "compiler_pair_state_consistency": 0.21875, "run": "main_expand_copy_seed456", "arm": "expand_copy", "seed": 456, "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "paired_heldout_L24", "n": 64, "max_steps": 24, "executor_accuracy": 0.328125, "executor_target_mass": 0.3278505615890026, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.9329427083333334, "arg_accuracy": 0.8899739583333334, "program_exact": 0.3125, "state_accuracy": 0.845703125, "state_all_exact": 0.3125, "state_prefix_fraction": 0.8450520833333334, "state_mean_correct_prefix": 20.28125, "executor_pair_answer_consistency": 0.03125, "executor_pair_both_correct": 0.03125, "compiler_pair_program_consistency": 0.03125, "compiler_pair_state_consistency": 0.03125, "run": "main_expand_copy_seed456", "arm": "expand_copy", "seed": 456, "global_step": 750, "stage_idx": 3 } ], "checkpoints": [ { "run": "main_expand_copy_seed456", "seed": 456, "tag": "stage1_max8_step00300", "stage": "stage1_max8", "global_step": 300, "compiler_max_steps": 8, "checkpoint_dir": "/workspace/large_artifacts/qwen_compiler_multiseed_reattribution/checkpoints/main_expand_copy_seed456/stage1_max8_step00300", "model_checkpoint": "/workspace/large_artifacts/qwen_compiler_multiseed_reattribution/checkpoints/main_expand_copy_seed456/stage1_max8_step00300/adapter", "heads_checkpoint": "/workspace/large_artifacts/qwen_compiler_multiseed_reattribution/checkpoints/main_expand_copy_seed456/stage1_max8_step00300/heads.pt", "selection_split": "paired_L8", "selection_metric": "executor_pair_both_correct", "selection_value": 1.0, "executor_accuracy": 1.0, "program_exact": 1.0, "state_prefix_fraction": 1.0 }, { "run": "main_expand_copy_seed456", "seed": 456, "tag": "stage2_max16_step00450", "stage": "stage2_max16", "global_step": 450, "compiler_max_steps": 16, "checkpoint_dir": "/workspace/large_artifacts/qwen_compiler_multiseed_reattribution/checkpoints/main_expand_copy_seed456/stage2_max16_step00450", "model_checkpoint": "/workspace/large_artifacts/qwen_compiler_multiseed_reattribution/checkpoints/main_expand_copy_seed456/stage2_max16_step00450/adapter", "heads_checkpoint": "/workspace/large_artifacts/qwen_compiler_multiseed_reattribution/checkpoints/main_expand_copy_seed456/stage2_max16_step00450/heads.pt", "selection_split": "paired_L8", "selection_metric": "executor_pair_both_correct", "selection_value": 1.0, "executor_accuracy": 1.0, "program_exact": 1.0, "state_prefix_fraction": 1.0 }, { "run": "main_expand_copy_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_expand_copy_seed456/stage3_max24_step00750", "model_checkpoint": "/workspace/large_artifacts/qwen_compiler_multiseed_reattribution/checkpoints/main_expand_copy_seed456/stage3_max24_step00750/adapter", "heads_checkpoint": "/workspace/large_artifacts/qwen_compiler_multiseed_reattribution/checkpoints/main_expand_copy_seed456/stage3_max24_step00750/heads.pt", "selection_split": "paired_L24", "selection_metric": "executor_pair_both_correct", "selection_value": 0.21875, "executor_accuracy": 0.546875, "program_exact": 0.546875, "state_prefix_fraction": 0.953125 } ], "elapsed_sec": 1914.379 }