{ "run": "main_expand_copy_seed789", "arm": "expand_copy", "seed": 789, "args": { "run_name": "main_expand_copy_seed789", "arm": "expand_copy", "model_id": "Qwen/Qwen3-4B", "seed": 789, "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_seed789", "arm": "expand_copy", "seed": 789, "stage": "stage1_max8", "stage_idx": 1, "local_step": 1, "global_step": 1, "compiler_max_steps": 8, "elapsed_sec": 2.546, "expansion_event": "init", "init_loss": 4.712630271911621, "op_loss": 1.1086543798446655, "arg_loss": 4.590086460113525, "state_loss": 4.578254699707031, "state_train_accuracy": 0.0, "executor_loss": 4.578107833862305, "loss": 47.47588348388672 }, { "run": "main_expand_copy_seed789", "arm": "expand_copy", "seed": 789, "stage": "stage1_max8", "stage_idx": 1, "local_step": 50, "global_step": 50, "compiler_max_steps": 8, "elapsed_sec": 56.875, "expansion_event": "init", "init_loss": 4.329229354858398, "op_loss": 1.209376335144043, "arg_loss": 3.620577335357666, "state_loss": 4.575301170349121, "state_train_accuracy": 0.0, "executor_loss": 4.573653221130371, "loss": 42.157562255859375 }, { "run": "main_expand_copy_seed789", "arm": "expand_copy", "seed": 789, "stage": "stage1_max8", "stage_idx": 1, "local_step": 100, "global_step": 100, "compiler_max_steps": 8, "elapsed_sec": 112.994, "expansion_event": "init", "init_loss": 3.166703939437866, "op_loss": 0.20880745351314545, "arg_loss": 1.5115083456039429, "state_loss": 4.2328572273254395, "state_train_accuracy": 0.0, "executor_loss": 4.161526203155518, "loss": 27.316038131713867 }, { "run": "main_expand_copy_seed789", "arm": "expand_copy", "seed": 789, "stage": "stage1_max8", "stage_idx": 1, "local_step": 150, "global_step": 150, "compiler_max_steps": 8, "elapsed_sec": 168.687, "expansion_event": "init", "init_loss": 1.5144336223602295, "op_loss": 0.00033540395088493824, "arg_loss": 0.04006514698266983, "state_loss": 1.7076807022094727, "state_train_accuracy": 0.5999999642372131, "executor_loss": 1.749406337738037, "loss": 9.675416946411133 }, { "run": "main_expand_copy_seed789", "arm": "expand_copy", "seed": 789, "stage": "stage1_max8", "stage_idx": 1, "local_step": 200, "global_step": 200, "compiler_max_steps": 8, "elapsed_sec": 224.86, "expansion_event": "init", "init_loss": 0.14735162258148193, "op_loss": 6.995269359322265e-05, "arg_loss": 0.0044110119342803955, "state_loss": 0.08465738594532013, "state_train_accuracy": 1.0, "executor_loss": 0.16748154163360596, "loss": 0.8592594265937805 }, { "run": "main_expand_copy_seed789", "arm": "expand_copy", "seed": 789, "stage": "stage1_max8", "stage_idx": 1, "local_step": 250, "global_step": 250, "compiler_max_steps": 8, "elapsed_sec": 280.947, "expansion_event": "init", "init_loss": 0.006545466370880604, "op_loss": 1.1347989129717462e-05, "arg_loss": 0.0006634441670030355, "state_loss": 0.008584892377257347, "state_train_accuracy": 1.0, "executor_loss": 0.009581603109836578, "loss": 0.04701348394155502 }, { "run": "main_expand_copy_seed789", "arm": "expand_copy", "seed": 789, "stage": "stage1_max8", "stage_idx": 1, "local_step": 300, "global_step": 300, "compiler_max_steps": 8, "elapsed_sec": 337.208, "expansion_event": "init", "init_loss": 0.013329234905540943, "op_loss": 2.8133347313996637e-06, "arg_loss": 0.0001475637691328302, "state_loss": 0.008137461729347706, "state_train_accuracy": 1.0, "executor_loss": 0.013893096707761288, "loss": 0.0759405642747879 }, { "run": "main_expand_copy_seed789", "arm": "expand_copy", "seed": 789, "stage": "stage2_max16", "stage_idx": 2, "local_step": 1, "global_step": 301, "compiler_max_steps": 16, "elapsed_sec": 357.516, "expansion_event": "expand_8_to_16_copy_last", "init_loss": 0.001358475536108017, "op_loss": 0.17388133704662323, "arg_loss": 0.8995564579963684, "state_loss": 0.808578610420227, "state_train_accuracy": 0.9230769276618958, "executor_loss": 2.601015567779541, "loss": 7.187135219573975 }, { "run": "main_expand_copy_seed789", "arm": "expand_copy", "seed": 789, "stage": "stage2_max16", "stage_idx": 2, "local_step": 50, "global_step": 350, "compiler_max_steps": 16, "elapsed_sec": 465.033, "expansion_event": "expand_8_to_16_copy_last", "init_loss": 0.0009786038426682353, "op_loss": 3.695487009736098e-07, "arg_loss": 5.8324611018178985e-05, "state_loss": 0.0012699533253908157, "state_train_accuracy": 1.0, "executor_loss": 0.001565553480759263, "loss": 0.006983590312302113 }, { "run": "main_expand_copy_seed789", "arm": "expand_copy", "seed": 789, "stage": "stage2_max16", "stage_idx": 2, "local_step": 100, "global_step": 400, "compiler_max_steps": 16, "elapsed_sec": 575.453, "expansion_event": "expand_8_to_16_copy_last", "init_loss": 0.000219175242818892, "op_loss": 2.3543830707239977e-07, "arg_loss": 1.631212762731593e-05, "state_loss": 0.00031767584732733667, "state_train_accuracy": 1.0, "executor_loss": 0.0003846988547593355, "loss": 0.001644559670239687 }, { "run": "main_expand_copy_seed789", "arm": "expand_copy", "seed": 789, "stage": "stage2_max16", "stage_idx": 2, "local_step": 150, "global_step": 450, "compiler_max_steps": 16, "elapsed_sec": 685.426, "expansion_event": "expand_8_to_16_copy_last", "init_loss": 0.00010281168215442449, "op_loss": 1.6483257070376567e-07, "arg_loss": 1.1254159289819654e-05, "state_loss": 0.0001735277910484001, "state_train_accuracy": 1.0, "executor_loss": 0.00021845567971467972, "loss": 0.0008484116988256574 }, { "run": "main_expand_copy_seed789", "arm": "expand_copy", "seed": 789, "stage": "stage3_max24", "stage_idx": 3, "local_step": 1, "global_step": 451, "compiler_max_steps": 24, "elapsed_sec": 749.303, "expansion_event": "expand_16_to_24_copy_last", "init_loss": 0.0001721650769468397, "op_loss": 0.432195782661438, "arg_loss": 0.5313873291015625, "state_loss": 0.7507462501525879, "state_train_accuracy": 0.9411764740943909, "executor_loss": 4.776092529296875, "loss": 8.085271835327148 }, { "run": "main_expand_copy_seed789", "arm": "expand_copy", "seed": 789, "stage": "stage3_max24", "stage_idx": 3, "local_step": 50, "global_step": 500, "compiler_max_steps": 24, "elapsed_sec": 918.889, "expansion_event": "expand_16_to_24_copy_last", "init_loss": 0.00014850308070890605, "op_loss": 0.0015125321224331856, "arg_loss": 0.0001772951945895329, "state_loss": 0.007801178842782974, "state_train_accuracy": 1.0, "executor_loss": 0.028427278622984886, "loss": 0.03904418274760246 }, { "run": "main_expand_copy_seed789", "arm": "expand_copy", "seed": 789, "stage": "stage3_max24", "stage_idx": 3, "local_step": 100, "global_step": 550, "compiler_max_steps": 24, "elapsed_sec": 1091.177, "expansion_event": "expand_16_to_24_copy_last", "init_loss": 0.00021337156067602336, "op_loss": 7.556188393209595e-06, "arg_loss": 0.00012920537847094238, "state_loss": 0.0012524421326816082, "state_train_accuracy": 1.0, "executor_loss": 0.0026914270129054785, "loss": 0.005321733187884092 }, { "run": "main_expand_copy_seed789", "arm": "expand_copy", "seed": 789, "stage": "stage3_max24", "stage_idx": 3, "local_step": 150, "global_step": 600, "compiler_max_steps": 24, "elapsed_sec": 1263.231, "expansion_event": "expand_16_to_24_copy_last", "init_loss": 0.0005368415731936693, "op_loss": 2.3236425477080047e-05, "arg_loss": 0.00018488321802578866, "state_loss": 0.0023986913729459047, "state_train_accuracy": 0.9999999403953552, "executor_loss": 0.0033931341022253036, "loss": 0.008701961487531662 }, { "run": "main_expand_copy_seed789", "arm": "expand_copy", "seed": 789, "stage": "stage3_max24", "stage_idx": 3, "local_step": 200, "global_step": 650, "compiler_max_steps": 24, "elapsed_sec": 1436.15, "expansion_event": "expand_16_to_24_copy_last", "init_loss": 0.05586610734462738, "op_loss": 0.2136288732290268, "arg_loss": 0.42742690443992615, "state_loss": 0.36198997497558594, "state_train_accuracy": 0.941605806350708, "executor_loss": 2.5312423706054688, "loss": 5.040033340454102 }, { "run": "main_expand_copy_seed789", "arm": "expand_copy", "seed": 789, "stage": "stage3_max24", "stage_idx": 3, "local_step": 250, "global_step": 700, "compiler_max_steps": 24, "elapsed_sec": 1607.522, "expansion_event": "expand_16_to_24_copy_last", "init_loss": 0.0008417941862717271, "op_loss": 3.123986243735999e-05, "arg_loss": 0.00018578016897663474, "state_loss": 0.002610813593491912, "state_train_accuracy": 1.0, "executor_loss": 0.003924151416867971, "loss": 0.01067650318145752 }, { "run": "main_expand_copy_seed789", "arm": "expand_copy", "seed": 789, "stage": "stage3_max24", "stage_idx": 3, "local_step": 300, "global_step": 750, "compiler_max_steps": 24, "elapsed_sec": 1780.0, "expansion_event": "expand_16_to_24_copy_last", "init_loss": 7.129862933652475e-05, "op_loss": 5.697456344933016e-06, "arg_loss": 6.0091020714025944e-05, "state_loss": 0.0006080219172872603, "state_train_accuracy": 1.0, "executor_loss": 0.0011898051016032696, "loss": 0.0023290831595659256 } ], "metrics": [ { "stage": "stage1_max8", "split": "standard_L8", "n": 64, "max_steps": 8, "executor_accuracy": 0.984375, "executor_target_mass": 0.9650754630565643, "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_seed789", "arm": "expand_copy", "seed": 789, "global_step": 300, "stage_idx": 1 }, { "stage": "stage1_max8", "split": "paraphrase_L8", "n": 64, "max_steps": 8, "executor_accuracy": 0.96875, "executor_target_mass": 0.9541471600532532, "direct_accuracy": NaN, "init_accuracy": 0.96875, "op_accuracy": 1.0, "arg_accuracy": 1.0, "program_exact": 0.96875, "state_accuracy": 0.96875, "state_all_exact": 0.96875, "state_prefix_fraction": 0.96875, "state_mean_correct_prefix": 7.75, "run": "main_expand_copy_seed789", "arm": "expand_copy", "seed": 789, "global_step": 300, "stage_idx": 1 }, { "stage": "stage1_max8", "split": "heldout_L8", "n": 64, "max_steps": 8, "executor_accuracy": 0.75, "executor_target_mass": 0.7403192594647408, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.953125, "arg_accuracy": 0.974609375, "program_exact": 0.75, "state_accuracy": 0.8515625, "state_all_exact": 0.75, "state_prefix_fraction": 0.8515625, "state_mean_correct_prefix": 6.8125, "run": "main_expand_copy_seed789", "arm": "expand_copy", "seed": 789, "global_step": 300, "stage_idx": 1 }, { "stage": "stage1_max8", "split": "paired_L8", "n": 64, "max_steps": 8, "executor_accuracy": 1.0, "executor_target_mass": 0.9915120452642441, "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_seed789", "arm": "expand_copy", "seed": 789, "global_step": 300, "stage_idx": 1 }, { "stage": "stage1_max8", "split": "paired_heldout_L8", "n": 64, "max_steps": 8, "executor_accuracy": 0.828125, "executor_target_mass": 0.8304917216300964, "direct_accuracy": NaN, "init_accuracy": 0.96875, "op_accuracy": 0.962890625, "arg_accuracy": 0.98046875, "program_exact": 0.8125, "state_accuracy": 0.892578125, "state_all_exact": 0.828125, "state_prefix_fraction": 0.892578125, "state_mean_correct_prefix": 7.140625, "executor_pair_answer_consistency": 0.71875, "executor_pair_both_correct": 0.6875, "compiler_pair_program_consistency": 0.6875, "compiler_pair_state_consistency": 0.71875, "run": "main_expand_copy_seed789", "arm": "expand_copy", "seed": 789, "global_step": 300, "stage_idx": 1 }, { "stage": "stage2_max16", "split": "standard_L8", "n": 64, "max_steps": 16, "executor_accuracy": 1.0, "executor_target_mass": 0.9888831004500389, "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_seed789", "arm": "expand_copy", "seed": 789, "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max16", "split": "paraphrase_L8", "n": 64, "max_steps": 16, "executor_accuracy": 1.0, "executor_target_mass": 0.9997521564364433, "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_seed789", "arm": "expand_copy", "seed": 789, "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max16", "split": "heldout_L8", "n": 64, "max_steps": 16, "executor_accuracy": 0.984375, "executor_target_mass": 0.9819936230778694, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.998046875, "arg_accuracy": 1.0, "program_exact": 0.984375, "state_accuracy": 0.998046875, "state_all_exact": 0.984375, "state_prefix_fraction": 0.998046875, "state_mean_correct_prefix": 7.984375, "run": "main_expand_copy_seed789", "arm": "expand_copy", "seed": 789, "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max16", "split": "paired_L8", "n": 64, "max_steps": 16, "executor_accuracy": 1.0, "executor_target_mass": 0.9997250139713287, "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_seed789", "arm": "expand_copy", "seed": 789, "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max16", "split": "paired_heldout_L8", "n": 64, "max_steps": 16, "executor_accuracy": 1.0, "executor_target_mass": 0.9924489930272102, "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_seed789", "arm": "expand_copy", "seed": 789, "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max16", "split": "standard_L16", "n": 64, "max_steps": 16, "executor_accuracy": 0.984375, "executor_target_mass": 0.9880921989679337, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 1.0, "arg_accuracy": 0.9990234375, "program_exact": 0.984375, "state_accuracy": 0.9970703125, "state_all_exact": 0.984375, "state_prefix_fraction": 0.9970703125, "state_mean_correct_prefix": 15.953125, "run": "main_expand_copy_seed789", "arm": "expand_copy", "seed": 789, "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max16", "split": "paraphrase_L16", "n": 64, "max_steps": 16, "executor_accuracy": 1.0, "executor_target_mass": 0.9995394349098206, "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_seed789", "arm": "expand_copy", "seed": 789, "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max16", "split": "heldout_L16", "n": 64, "max_steps": 16, "executor_accuracy": 0.90625, "executor_target_mass": 0.8662520945072174, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.9794921875, "arg_accuracy": 0.970703125, "program_exact": 0.90625, "state_accuracy": 0.9560546875, "state_all_exact": 0.90625, "state_prefix_fraction": 0.955078125, "state_mean_correct_prefix": 15.28125, "run": "main_expand_copy_seed789", "arm": "expand_copy", "seed": 789, "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max16", "split": "paired_L16", "n": 64, "max_steps": 16, "executor_accuracy": 0.984375, "executor_target_mass": 0.984925277531147, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 1.0, "arg_accuracy": 0.9990234375, "program_exact": 0.984375, "state_accuracy": 0.998046875, "state_all_exact": 0.984375, "state_prefix_fraction": 0.998046875, "state_mean_correct_prefix": 15.96875, "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_expand_copy_seed789", "arm": "expand_copy", "seed": 789, "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max16", "split": "paired_heldout_L16", "n": 64, "max_steps": 16, "executor_accuracy": 0.96875, "executor_target_mass": 0.9395963996648788, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.994140625, "arg_accuracy": 0.9873046875, "program_exact": 0.96875, "state_accuracy": 0.9833984375, "state_all_exact": 0.96875, "state_prefix_fraction": 0.9833984375, "state_mean_correct_prefix": 15.734375, "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_seed789", "arm": "expand_copy", "seed": 789, "global_step": 450, "stage_idx": 2 }, { "stage": "stage3_max24", "split": "standard_L8", "n": 64, "max_steps": 24, "executor_accuracy": 0.984375, "executor_target_mass": 0.9813942387700081, "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_seed789", "arm": "expand_copy", "seed": 789, "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "paraphrase_L8", "n": 64, "max_steps": 24, "executor_accuracy": 0.984375, "executor_target_mass": 0.9822830110788345, "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_seed789", "arm": "expand_copy", "seed": 789, "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "heldout_L8", "n": 64, "max_steps": 24, "executor_accuracy": 0.921875, "executor_target_mass": 0.9230572804808617, "direct_accuracy": NaN, "init_accuracy": 0.96875, "op_accuracy": 0.98828125, "arg_accuracy": 0.984375, "program_exact": 0.921875, "state_accuracy": 0.9453125, "state_all_exact": 0.921875, "state_prefix_fraction": 0.9453125, "state_mean_correct_prefix": 7.5625, "run": "main_expand_copy_seed789", "arm": "expand_copy", "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": 0.9934751987457275, "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_seed789", "arm": "expand_copy", "seed": 789, "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": 0.971853956580162, "direct_accuracy": NaN, "init_accuracy": 0.96875, "op_accuracy": 1.0, "arg_accuracy": 1.0, "program_exact": 0.96875, "state_accuracy": 0.96875, "state_all_exact": 0.96875, "state_prefix_fraction": 0.96875, "state_mean_correct_prefix": 7.75, "executor_pair_answer_consistency": 1.0, "executor_pair_both_correct": 0.96875, "compiler_pair_program_consistency": 1.0, "compiler_pair_state_consistency": 1.0, "run": "main_expand_copy_seed789", "arm": "expand_copy", "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": 0.9987399354577065, "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_seed789", "arm": "expand_copy", "seed": 789, "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "paraphrase_L16", "n": 64, "max_steps": 24, "executor_accuracy": 0.984375, "executor_target_mass": 0.9849391207098961, "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_seed789", "arm": "expand_copy", "seed": 789, "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "heldout_L16", "n": 64, "max_steps": 24, "executor_accuracy": 0.859375, "executor_target_mass": 0.8617495149374008, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.9892578125, "arg_accuracy": 0.9873046875, "program_exact": 0.859375, "state_accuracy": 0.896484375, "state_all_exact": 0.859375, "state_prefix_fraction": 0.896484375, "state_mean_correct_prefix": 14.34375, "run": "main_expand_copy_seed789", "arm": "expand_copy", "seed": 789, "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "paired_L16", "n": 64, "max_steps": 24, "executor_accuracy": 0.96875, "executor_target_mass": 0.9697678908705711, "direct_accuracy": NaN, "init_accuracy": 0.96875, "op_accuracy": 1.0, "arg_accuracy": 1.0, "program_exact": 0.96875, "state_accuracy": 0.96875, "state_all_exact": 0.96875, "state_prefix_fraction": 0.96875, "state_mean_correct_prefix": 15.5, "executor_pair_answer_consistency": 1.0, "executor_pair_both_correct": 0.96875, "compiler_pair_program_consistency": 1.0, "compiler_pair_state_consistency": 1.0, "run": "main_expand_copy_seed789", "arm": "expand_copy", "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": 0.9978818520903587, "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_seed789", "arm": "expand_copy", "seed": 789, "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "standard_L24", "n": 64, "max_steps": 24, "executor_accuracy": 0.8125, "executor_target_mass": 0.8227786868810654, "direct_accuracy": NaN, "init_accuracy": 0.96875, "op_accuracy": 0.9986979166666666, "arg_accuracy": 0.9928385416666666, "program_exact": 0.8125, "state_accuracy": 0.94921875, "state_all_exact": 0.8125, "state_prefix_fraction": 0.94921875, "state_mean_correct_prefix": 22.78125, "run": "main_expand_copy_seed789", "arm": "expand_copy", "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": 0.982705220580101, "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": 23.625, "run": "main_expand_copy_seed789", "arm": "expand_copy", "seed": 789, "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "heldout_L24", "n": 64, "max_steps": 24, "executor_accuracy": 0.765625, "executor_target_mass": 0.7613858543336391, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.9811197916666666, "arg_accuracy": 0.9765625, "program_exact": 0.765625, "state_accuracy": 0.9283854166666666, "state_all_exact": 0.765625, "state_prefix_fraction": 0.9283854166666666, "state_mean_correct_prefix": 22.28125, "run": "main_expand_copy_seed789", "arm": "expand_copy", "seed": 789, "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "paired_L24", "n": 64, "max_steps": 24, "executor_accuracy": 0.9375, "executor_target_mass": 0.9385778158903122, "direct_accuracy": NaN, "init_accuracy": 0.96875, "op_accuracy": 1.0, "arg_accuracy": 0.9986979166666666, "program_exact": 0.9375, "state_accuracy": 0.9661458333333334, "state_all_exact": 0.9375, "state_prefix_fraction": 0.9661458333333334, "state_mean_correct_prefix": 23.1875, "executor_pair_answer_consistency": 0.9375, "executor_pair_both_correct": 0.90625, "compiler_pair_program_consistency": 0.9375, "compiler_pair_state_consistency": 0.9375, "run": "main_expand_copy_seed789", "arm": "expand_copy", "seed": 789, "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "paired_heldout_L24", "n": 64, "max_steps": 24, "executor_accuracy": 0.796875, "executor_target_mass": 0.7651928812265396, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.9869791666666666, "arg_accuracy": 0.9830729166666666, "program_exact": 0.796875, "state_accuracy": 0.95703125, "state_all_exact": 0.796875, "state_prefix_fraction": 0.9563802083333334, "state_mean_correct_prefix": 22.953125, "executor_pair_answer_consistency": 0.59375, "executor_pair_both_correct": 0.59375, "compiler_pair_program_consistency": 0.59375, "compiler_pair_state_consistency": 0.59375, "run": "main_expand_copy_seed789", "arm": "expand_copy", "seed": 789, "global_step": 750, "stage_idx": 3 } ], "checkpoints": [ { "run": "main_expand_copy_seed789", "seed": 789, "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_seed789/stage1_max8_step00300", "model_checkpoint": "/workspace/large_artifacts/qwen_compiler_multiseed_reattribution/checkpoints/main_expand_copy_seed789/stage1_max8_step00300/adapter", "heads_checkpoint": "/workspace/large_artifacts/qwen_compiler_multiseed_reattribution/checkpoints/main_expand_copy_seed789/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_seed789", "seed": 789, "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_seed789/stage2_max16_step00450", "model_checkpoint": "/workspace/large_artifacts/qwen_compiler_multiseed_reattribution/checkpoints/main_expand_copy_seed789/stage2_max16_step00450/adapter", "heads_checkpoint": "/workspace/large_artifacts/qwen_compiler_multiseed_reattribution/checkpoints/main_expand_copy_seed789/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_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_expand_copy_seed789/stage3_max24_step00750", "model_checkpoint": "/workspace/large_artifacts/qwen_compiler_multiseed_reattribution/checkpoints/main_expand_copy_seed789/stage3_max24_step00750/adapter", "heads_checkpoint": "/workspace/large_artifacts/qwen_compiler_multiseed_reattribution/checkpoints/main_expand_copy_seed789/stage3_max24_step00750/heads.pt", "selection_split": "paired_L24", "selection_metric": "executor_pair_both_correct", "selection_value": 0.90625, "executor_accuracy": 0.9375, "program_exact": 0.9375, "state_prefix_fraction": 0.9661458333333334 } ], "elapsed_sec": 1907.543 }