{ "run": "main_expand_copy_seed123", "arm": "expand_copy", "seed": 123, "args": { "run_name": "main_expand_copy_seed123", "arm": "expand_copy", "model_id": "Qwen/Qwen3-4B", "seed": 123, "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_seed123", "arm": "expand_copy", "seed": 123, "stage": "stage1_max8", "stage_idx": 1, "local_step": 1, "global_step": 1, "compiler_max_steps": 8, "elapsed_sec": 2.621, "expansion_event": "init", "init_loss": 4.664542198181152, "op_loss": 1.146849274635315, "arg_loss": 4.618834972381592, "state_loss": 4.577921390533447, "state_train_accuracy": 0.0, "executor_loss": 4.577940940856934, "loss": 47.436222076416016 }, { "run": "main_expand_copy_seed123", "arm": "expand_copy", "seed": 123, "stage": "stage1_max8", "stage_idx": 1, "local_step": 50, "global_step": 50, "compiler_max_steps": 8, "elapsed_sec": 56.221, "expansion_event": "init", "init_loss": 4.519749641418457, "op_loss": 0.8716250061988831, "arg_loss": 3.662632465362549, "state_loss": 4.569859504699707, "state_train_accuracy": 0.0, "executor_loss": 4.576577663421631, "loss": 42.747589111328125 }, { "run": "main_expand_copy_seed123", "arm": "expand_copy", "seed": 123, "stage": "stage1_max8", "stage_idx": 1, "local_step": 100, "global_step": 100, "compiler_max_steps": 8, "elapsed_sec": 111.543, "expansion_event": "init", "init_loss": 2.1171388626098633, "op_loss": 0.013011542148888111, "arg_loss": 1.2907854318618774, "state_loss": 3.4374237060546875, "state_train_accuracy": 0.6428571939468384, "executor_loss": 3.8883216381073, "loss": 20.9704532623291 }, { "run": "main_expand_copy_seed123", "arm": "expand_copy", "seed": 123, "stage": "stage1_max8", "stage_idx": 1, "local_step": 150, "global_step": 150, "compiler_max_steps": 8, "elapsed_sec": 167.486, "expansion_event": "init", "init_loss": 0.30911803245544434, "op_loss": 0.00021766196005046368, "arg_loss": 0.01309637539088726, "state_loss": 0.2872236371040344, "state_train_accuracy": 1.0, "executor_loss": 0.3835184574127197, "loss": 1.9598174095153809 }, { "run": "main_expand_copy_seed123", "arm": "expand_copy", "seed": 123, "stage": "stage1_max8", "stage_idx": 1, "local_step": 200, "global_step": 200, "compiler_max_steps": 8, "elapsed_sec": 223.308, "expansion_event": "init", "init_loss": 0.010477048344910145, "op_loss": 1.430497104593087e-05, "arg_loss": 0.0010767225176095963, "state_loss": 0.015372047200798988, "state_train_accuracy": 1.0, "executor_loss": 0.017158249393105507, "loss": 0.07875968515872955 }, { "run": "main_expand_copy_seed123", "arm": "expand_copy", "seed": 123, "stage": "stage1_max8", "stage_idx": 1, "local_step": 250, "global_step": 250, "compiler_max_steps": 8, "elapsed_sec": 279.16, "expansion_event": "init", "init_loss": 0.4113067388534546, "op_loss": 3.875557013088837e-06, "arg_loss": 0.00022363022435456514, "state_loss": 0.5595079660415649, "state_train_accuracy": 0.8297871947288513, "executor_loss": 0.41259026527404785, "loss": 2.6182236671447754 }, { "run": "main_expand_copy_seed123", "arm": "expand_copy", "seed": 123, "stage": "stage1_max8", "stage_idx": 1, "local_step": 300, "global_step": 300, "compiler_max_steps": 8, "elapsed_sec": 335.38, "expansion_event": "init", "init_loss": 0.0010023314971476793, "op_loss": 1.7994907466345467e-06, "arg_loss": 0.00011582831939449534, "state_loss": 0.001409538439475, "state_train_accuracy": 1.0, "executor_loss": 0.0016197934746742249, "loss": 0.007503770757466555 }, { "run": "main_expand_copy_seed123", "arm": "expand_copy", "seed": 123, "stage": "stage2_max16", "stage_idx": 2, "local_step": 1, "global_step": 301, "compiler_max_steps": 16, "elapsed_sec": 355.896, "expansion_event": "expand_8_to_16_copy_last", "init_loss": 0.0007871515117585659, "op_loss": 0.4094744026660919, "arg_loss": 0.2865528166294098, "state_loss": 0.47255316376686096, "state_train_accuracy": 0.9534883499145508, "executor_loss": 1.2495537996292114, "loss": 3.2809414863586426 }, { "run": "main_expand_copy_seed123", "arm": "expand_copy", "seed": 123, "stage": "stage2_max16", "stage_idx": 2, "local_step": 50, "global_step": 350, "compiler_max_steps": 16, "elapsed_sec": 463.584, "expansion_event": "expand_8_to_16_copy_last", "init_loss": 0.001004744553938508, "op_loss": 4.115951924177352e-06, "arg_loss": 0.00010258788097416982, "state_loss": 0.0017706583021208644, "state_train_accuracy": 0.9999999403953552, "executor_loss": 0.0017382700461894274, "loss": 0.007942373864352703 }, { "run": "main_expand_copy_seed123", "arm": "expand_copy", "seed": 123, "stage": "stage2_max16", "stage_idx": 2, "local_step": 100, "global_step": 400, "compiler_max_steps": 16, "elapsed_sec": 573.153, "expansion_event": "expand_8_to_16_copy_last", "init_loss": 0.00025157813797704875, "op_loss": 1.4037908613317995e-06, "arg_loss": 2.895287616411224e-05, "state_loss": 0.00043421483132988214, "state_train_accuracy": 1.0, "executor_loss": 0.0004715698305517435, "loss": 0.0020293123088777065 }, { "run": "main_expand_copy_seed123", "arm": "expand_copy", "seed": 123, "stage": "stage2_max16", "stage_idx": 2, "local_step": 150, "global_step": 450, "compiler_max_steps": 16, "elapsed_sec": 682.974, "expansion_event": "expand_8_to_16_copy_last", "init_loss": 0.00011836659541586414, "op_loss": 6.239858976186952e-07, "arg_loss": 1.4645829651271924e-05, "state_loss": 0.00019638767116703093, "state_train_accuracy": 1.0, "executor_loss": 0.0002395085321040824, "loss": 0.00096856988966465 }, { "run": "main_expand_copy_seed123", "arm": "expand_copy", "seed": 123, "stage": "stage3_max24", "stage_idx": 3, "local_step": 1, "global_step": 451, "compiler_max_steps": 24, "elapsed_sec": 746.708, "expansion_event": "expand_16_to_24_copy_last", "init_loss": 0.00014893722254782915, "op_loss": 0.2647160589694977, "arg_loss": 0.15986043214797974, "state_loss": 0.28388887643814087, "state_train_accuracy": 0.9428572058677673, "executor_loss": 1.522825002670288, "loss": 2.7114672660827637 }, { "run": "main_expand_copy_seed123", "arm": "expand_copy", "seed": 123, "stage": "stage3_max24", "stage_idx": 3, "local_step": 50, "global_step": 500, "compiler_max_steps": 24, "elapsed_sec": 916.879, "expansion_event": "expand_16_to_24_copy_last", "init_loss": 0.00011020118836313486, "op_loss": 1.4024788470123895e-05, "arg_loss": 0.0004962439415976405, "state_loss": 0.004077180754393339, "state_train_accuracy": 1.0, "executor_loss": 0.007761813700199127, "loss": 0.014278799295425415 }, { "run": "main_expand_copy_seed123", "arm": "expand_copy", "seed": 123, "stage": "stage3_max24", "stage_idx": 3, "local_step": 100, "global_step": 550, "compiler_max_steps": 24, "elapsed_sec": 1090.557, "expansion_event": "expand_16_to_24_copy_last", "init_loss": 6.170335836941376e-05, "op_loss": 0.2872176766395569, "arg_loss": 0.8955934047698975, "state_loss": 1.2286409139633179, "state_train_accuracy": 0.7633587718009949, "executor_loss": 2.353351593017578, "loss": 7.451830863952637 }, { "run": "main_expand_copy_seed123", "arm": "expand_copy", "seed": 123, "stage": "stage3_max24", "stage_idx": 3, "local_step": 150, "global_step": 600, "compiler_max_steps": 24, "elapsed_sec": 1263.958, "expansion_event": "expand_16_to_24_copy_last", "init_loss": 3.017416747752577e-05, "op_loss": 0.09283832460641861, "arg_loss": 0.11019756644964218, "state_loss": 0.17161217331886292, "state_train_accuracy": 0.985714316368103, "executor_loss": 1.4841152429580688, "loss": 2.189476728439331 }, { "run": "main_expand_copy_seed123", "arm": "expand_copy", "seed": 123, "stage": "stage3_max24", "stage_idx": 3, "local_step": 200, "global_step": 650, "compiler_max_steps": 24, "elapsed_sec": 1437.831, "expansion_event": "expand_16_to_24_copy_last", "init_loss": 1.5660980352549814e-05, "op_loss": 2.286881226609694e-06, "arg_loss": 2.2530040951096453e-05, "state_loss": 0.00022562923550140113, "state_train_accuracy": 1.0, "executor_loss": 0.00043748534517362714, "loss": 0.0008181655430234969 }, { "run": "main_expand_copy_seed123", "arm": "expand_copy", "seed": 123, "stage": "stage3_max24", "stage_idx": 3, "local_step": 250, "global_step": 700, "compiler_max_steps": 24, "elapsed_sec": 1611.275, "expansion_event": "expand_16_to_24_copy_last", "init_loss": 3.959138848586008e-05, "op_loss": 0.0004436022136360407, "arg_loss": 0.0011440373491495848, "state_loss": 0.0058250995352864265, "state_train_accuracy": 1.0, "executor_loss": 0.024642735719680786, "loss": 0.035645950585603714 }, { "run": "main_expand_copy_seed123", "arm": "expand_copy", "seed": 123, "stage": "stage3_max24", "stage_idx": 3, "local_step": 300, "global_step": 750, "compiler_max_steps": 24, "elapsed_sec": 1784.679, "expansion_event": "expand_16_to_24_copy_last", "init_loss": 0.010943728499114513, "op_loss": 3.335912879265379e-06, "arg_loss": 4.678138793678954e-05, "state_loss": 0.016725528985261917, "state_train_accuracy": 0.9999999403953552, "executor_loss": 0.011714239604771137, "loss": 0.072405144572258 } ], "metrics": [ { "stage": "stage1_max8", "split": "standard_L8", "n": 64, "max_steps": 8, "executor_accuracy": 0.984375, "executor_target_mass": 0.9632068425416946, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 1.0, "arg_accuracy": 0.998046875, "program_exact": 0.984375, "state_accuracy": 0.990234375, "state_all_exact": 0.984375, "state_prefix_fraction": 0.990234375, "state_mean_correct_prefix": 7.921875, "run": "main_expand_copy_seed123", "arm": "expand_copy", "seed": 123, "global_step": 300, "stage_idx": 1 }, { "stage": "stage1_max8", "split": "paraphrase_L8", "n": 64, "max_steps": 8, "executor_accuracy": 0.984375, "executor_target_mass": 0.9669913426041603, "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_seed123", "arm": "expand_copy", "seed": 123, "global_step": 300, "stage_idx": 1 }, { "stage": "stage1_max8", "split": "heldout_L8", "n": 64, "max_steps": 8, "executor_accuracy": 0.734375, "executor_target_mass": 0.7142626643180847, "direct_accuracy": NaN, "init_accuracy": 0.96875, "op_accuracy": 0.939453125, "arg_accuracy": 0.9453125, "program_exact": 0.734375, "state_accuracy": 0.826171875, "state_all_exact": 0.734375, "state_prefix_fraction": 0.826171875, "state_mean_correct_prefix": 6.609375, "run": "main_expand_copy_seed123", "arm": "expand_copy", "seed": 123, "global_step": 300, "stage_idx": 1 }, { "stage": "stage1_max8", "split": "paired_L8", "n": 64, "max_steps": 8, "executor_accuracy": 0.96875, "executor_target_mass": 0.9664822518825531, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.99609375, "arg_accuracy": 0.99609375, "program_exact": 0.96875, "state_accuracy": 0.98046875, "state_all_exact": 0.96875, "state_prefix_fraction": 0.98046875, "state_mean_correct_prefix": 7.84375, "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_seed123", "arm": "expand_copy", "seed": 123, "global_step": 300, "stage_idx": 1 }, { "stage": "stage1_max8", "split": "paired_heldout_L8", "n": 64, "max_steps": 8, "executor_accuracy": 0.875, "executor_target_mass": 0.8698408156633377, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.96875, "arg_accuracy": 0.958984375, "program_exact": 0.875, "state_accuracy": 0.93359375, "state_all_exact": 0.875, "state_prefix_fraction": 0.93359375, "state_mean_correct_prefix": 7.46875, "executor_pair_answer_consistency": 0.75, "executor_pair_both_correct": 0.75, "compiler_pair_program_consistency": 0.75, "compiler_pair_state_consistency": 0.75, "run": "main_expand_copy_seed123", "arm": "expand_copy", "seed": 123, "global_step": 300, "stage_idx": 1 }, { "stage": "stage2_max16", "split": "standard_L8", "n": 64, "max_steps": 16, "executor_accuracy": 1.0, "executor_target_mass": 0.9987975507974625, "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_seed123", "arm": "expand_copy", "seed": 123, "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max16", "split": "paraphrase_L8", "n": 64, "max_steps": 16, "executor_accuracy": 1.0, "executor_target_mass": 0.9997294098138809, "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_seed123", "arm": "expand_copy", "seed": 123, "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max16", "split": "heldout_L8", "n": 64, "max_steps": 16, "executor_accuracy": 0.96875, "executor_target_mass": 0.9678959548473358, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.986328125, "arg_accuracy": 0.978515625, "program_exact": 0.96875, "state_accuracy": 0.978515625, "state_all_exact": 0.96875, "state_prefix_fraction": 0.978515625, "state_mean_correct_prefix": 7.828125, "run": "main_expand_copy_seed123", "arm": "expand_copy", "seed": 123, "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max16", "split": "paired_L8", "n": 64, "max_steps": 16, "executor_accuracy": 1.0, "executor_target_mass": 0.999403603374958, "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_seed123", "arm": "expand_copy", "seed": 123, "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": 0.9931009039282799, "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_seed123", "arm": "expand_copy", "seed": 123, "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max16", "split": "standard_L16", "n": 64, "max_steps": 16, "executor_accuracy": 0.984375, "executor_target_mass": 0.9841509982943535, "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, "run": "main_expand_copy_seed123", "arm": "expand_copy", "seed": 123, "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max16", "split": "paraphrase_L16", "n": 64, "max_steps": 16, "executor_accuracy": 1.0, "executor_target_mass": 0.9995570406317711, "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_seed123", "arm": "expand_copy", "seed": 123, "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max16", "split": "heldout_L16", "n": 64, "max_steps": 16, "executor_accuracy": 0.796875, "executor_target_mass": 0.7887180894613266, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.9833984375, "arg_accuracy": 0.9765625, "program_exact": 0.796875, "state_accuracy": 0.953125, "state_all_exact": 0.796875, "state_prefix_fraction": 0.953125, "state_mean_correct_prefix": 15.25, "run": "main_expand_copy_seed123", "arm": "expand_copy", "seed": 123, "global_step": 450, "stage_idx": 2 }, { "stage": "stage2_max16", "split": "paired_L16", "n": 64, "max_steps": 16, "executor_accuracy": 1.0, "executor_target_mass": 0.9994222670793533, "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_seed123", "arm": "expand_copy", "seed": 123, "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": 0.9211421236395836, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.9853515625, "arg_accuracy": 0.98046875, "program_exact": 0.921875, "state_accuracy": 0.9765625, "state_all_exact": 0.921875, "state_prefix_fraction": 0.9765625, "state_mean_correct_prefix": 15.625, "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_seed123", "arm": "expand_copy", "seed": 123, "global_step": 450, "stage_idx": 2 }, { "stage": "stage3_max24", "split": "standard_L8", "n": 64, "max_steps": 24, "executor_accuracy": 1.0, "executor_target_mass": 0.9978929162025452, "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_seed123", "arm": "expand_copy", "seed": 123, "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "paraphrase_L8", "n": 64, "max_steps": 24, "executor_accuracy": 1.0, "executor_target_mass": 0.9993784055113792, "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_seed123", "arm": "expand_copy", "seed": 123, "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "heldout_L8", "n": 64, "max_steps": 24, "executor_accuracy": 0.46875, "executor_target_mass": 0.4817279428243637, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.873046875, "arg_accuracy": 0.9296875, "program_exact": 0.46875, "state_accuracy": 0.625, "state_all_exact": 0.46875, "state_prefix_fraction": 0.62109375, "state_mean_correct_prefix": 4.96875, "run": "main_expand_copy_seed123", "arm": "expand_copy", "seed": 123, "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "paired_L8", "n": 64, "max_steps": 24, "executor_accuracy": 1.0, "executor_target_mass": 0.9996174722909927, "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_seed123", "arm": "expand_copy", "seed": 123, "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "paired_heldout_L8", "n": 64, "max_steps": 24, "executor_accuracy": 0.75, "executor_target_mass": 0.7476436123251915, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.93359375, "arg_accuracy": 0.958984375, "program_exact": 0.75, "state_accuracy": 0.83203125, "state_all_exact": 0.75, "state_prefix_fraction": 0.83203125, "state_mean_correct_prefix": 6.65625, "executor_pair_answer_consistency": 0.5, "executor_pair_both_correct": 0.5, "compiler_pair_program_consistency": 0.5, "compiler_pair_state_consistency": 0.5, "run": "main_expand_copy_seed123", "arm": "expand_copy", "seed": 123, "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "standard_L16", "n": 64, "max_steps": 24, "executor_accuracy": 1.0, "executor_target_mass": 0.9990312084555626, "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_seed123", "arm": "expand_copy", "seed": 123, "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "paraphrase_L16", "n": 64, "max_steps": 24, "executor_accuracy": 1.0, "executor_target_mass": 0.999056451022625, "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_seed123", "arm": "expand_copy", "seed": 123, "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "heldout_L16", "n": 64, "max_steps": 24, "executor_accuracy": 0.359375, "executor_target_mass": 0.3768988847732544, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.8720703125, "arg_accuracy": 0.9130859375, "program_exact": 0.359375, "state_accuracy": 0.5185546875, "state_all_exact": 0.359375, "state_prefix_fraction": 0.5185546875, "state_mean_correct_prefix": 8.296875, "run": "main_expand_copy_seed123", "arm": "expand_copy", "seed": 123, "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "paired_L16", "n": 64, "max_steps": 24, "executor_accuracy": 1.0, "executor_target_mass": 0.9992646798491478, "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_seed123", "arm": "expand_copy", "seed": 123, "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "paired_heldout_L16", "n": 64, "max_steps": 24, "executor_accuracy": 0.6875, "executor_target_mass": 0.6755928173661232, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.9375, "arg_accuracy": 0.9462890625, "program_exact": 0.671875, "state_accuracy": 0.8046875, "state_all_exact": 0.671875, "state_prefix_fraction": 0.8017578125, "state_mean_correct_prefix": 12.828125, "executor_pair_answer_consistency": 0.375, "executor_pair_both_correct": 0.375, "compiler_pair_program_consistency": 0.34375, "compiler_pair_state_consistency": 0.34375, "run": "main_expand_copy_seed123", "arm": "expand_copy", "seed": 123, "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "standard_L24", "n": 64, "max_steps": 24, "executor_accuracy": 0.0, "executor_target_mass": 0.0006909116864335374, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.798828125, "arg_accuracy": 0.73046875, "program_exact": 0.0, "state_accuracy": 0.720703125, "state_all_exact": 0.0, "state_prefix_fraction": 0.7174479166666666, "state_mean_correct_prefix": 17.21875, "run": "main_expand_copy_seed123", "arm": "expand_copy", "seed": 123, "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "paraphrase_L24", "n": 64, "max_steps": 24, "executor_accuracy": 0.78125, "executor_target_mass": 0.7721076384186745, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.98046875, "arg_accuracy": 0.97265625, "program_exact": 0.71875, "state_accuracy": 0.9752604166666666, "state_all_exact": 0.71875, "state_prefix_fraction": 0.97265625, "state_mean_correct_prefix": 23.34375, "run": "main_expand_copy_seed123", "arm": "expand_copy", "seed": 123, "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "heldout_L24", "n": 64, "max_steps": 24, "executor_accuracy": 0.1875, "executor_target_mass": 0.1772532444447279, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.8587239583333334, "arg_accuracy": 0.869140625, "program_exact": 0.125, "state_accuracy": 0.4544270833333333, "state_all_exact": 0.125, "state_prefix_fraction": 0.4368489583333333, "state_mean_correct_prefix": 10.484375, "run": "main_expand_copy_seed123", "arm": "expand_copy", "seed": 123, "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "paired_L24", "n": 64, "max_steps": 24, "executor_accuracy": 0.328125, "executor_target_mass": 0.3270937381312251, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.904296875, "arg_accuracy": 0.8567708333333334, "program_exact": 0.3125, "state_accuracy": 0.8489583333333334, "state_all_exact": 0.3125, "state_prefix_fraction": 0.8463541666666666, "state_mean_correct_prefix": 20.3125, "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_expand_copy_seed123", "arm": "expand_copy", "seed": 123, "global_step": 750, "stage_idx": 3 }, { "stage": "stage3_max24", "split": "paired_heldout_L24", "n": 64, "max_steps": 24, "executor_accuracy": 0.09375, "executor_target_mass": 0.0982597434121999, "direct_accuracy": NaN, "init_accuracy": 1.0, "op_accuracy": 0.8294270833333334, "arg_accuracy": 0.7799479166666666, "program_exact": 0.0625, "state_accuracy": 0.5546875, "state_all_exact": 0.0625, "state_prefix_fraction": 0.5494791666666666, "state_mean_correct_prefix": 13.1875, "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_expand_copy_seed123", "arm": "expand_copy", "seed": 123, "global_step": 750, "stage_idx": 3 } ], "checkpoints": [ { "run": "main_expand_copy_seed123", "seed": 123, "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_seed123/stage1_max8_step00300", "model_checkpoint": "/workspace/large_artifacts/qwen_compiler_multiseed_reattribution/checkpoints/main_expand_copy_seed123/stage1_max8_step00300/adapter", "heads_checkpoint": "/workspace/large_artifacts/qwen_compiler_multiseed_reattribution/checkpoints/main_expand_copy_seed123/stage1_max8_step00300/heads.pt", "selection_split": "paired_L8", "selection_metric": "executor_pair_both_correct", "selection_value": 0.9375, "executor_accuracy": 0.96875, "program_exact": 0.96875, "state_prefix_fraction": 0.98046875 }, { "run": "main_expand_copy_seed123", "seed": 123, "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_seed123/stage2_max16_step00450", "model_checkpoint": "/workspace/large_artifacts/qwen_compiler_multiseed_reattribution/checkpoints/main_expand_copy_seed123/stage2_max16_step00450/adapter", "heads_checkpoint": "/workspace/large_artifacts/qwen_compiler_multiseed_reattribution/checkpoints/main_expand_copy_seed123/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_seed123", "seed": 123, "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_seed123/stage3_max24_step00750", "model_checkpoint": "/workspace/large_artifacts/qwen_compiler_multiseed_reattribution/checkpoints/main_expand_copy_seed123/stage3_max24_step00750/adapter", "heads_checkpoint": "/workspace/large_artifacts/qwen_compiler_multiseed_reattribution/checkpoints/main_expand_copy_seed123/stage3_max24_step00750/heads.pt", "selection_split": "paired_L24", "selection_metric": "executor_pair_both_correct", "selection_value": 0.0, "executor_accuracy": 0.328125, "program_exact": 0.3125, "state_prefix_fraction": 0.8463541666666666 } ], "elapsed_sec": 1912.101 }