{ "args": { "model_id": "Qwen/Qwen3.5-4B", "text_model_fallback": "Qwen/Qwen3-4B", "model_class": "auto", "load_in_4bit": true, "torch_dtype": "bf16", "device_map": "auto", "modulus": 97, "max_steps": 12, "train_min_len": 1, "train_max_len": 4, "eval_lengths": "4,8,12", "train_size": 1024, "eval_size": 256, "feature_batch_size": 8, "train_batch_size": 128, "eval_batch_size": 256, "train_steps": 800, "eval_every": 200, "lr": 0.001, "weight_decay": 0.01, "grad_clip": 1.0, "head_width": 512, "trace_loss_weight": 1.0, "executor_loss_weight": 1.0, "variants": "direct,compiler_trace,compiler_answer_only", "max_length": 512, "seed": 7, "output_dir": "experiments/qwen_structured_bridge/runs/main_qwen35_numeric_spans", "checkpoint_dir": "", "verbose": false }, "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": "AutoModelForMultimodalLM", "peft_installed": true, "gpu_name": "NVIDIA RTX 6000 Ada Generation", "gpu_vram_gb": 47.383 }, "hidden_dim": 2560, "dataset": { "train_size": 1024, "eval_size": 256, "train_lengths": [ 1, 4 ], "eval_lengths": [ 4, 8, 12 ] }, "variants": { "direct": { "variant": "direct", "train_seconds": 24.01050329208374, "checkpoints": [ "large_artifacts/qwen_structured_bridge/checkpoints/main_qwen35_numeric_spans/direct/checkpoint_final.pt" ], "train_log": [ { "step": 1, "loss": 4.59930419921875, "direct_loss": 4.59930419921875, "len4_direct_accuracy": 0.0078125, "len4_direct_target_mass": 0.00916213821619749, "len4_executor_accuracy": NaN, "len4_executor_target_mass": NaN, "len4_init_accuracy": NaN, "len4_op_accuracy": NaN, "len4_arg_accuracy": NaN, "len4_program_exact": NaN, "len8_direct_accuracy": 0.01171875, "len8_direct_target_mass": 0.01064371969550848, "len8_executor_accuracy": NaN, "len8_executor_target_mass": NaN, "len8_init_accuracy": NaN, "len8_op_accuracy": NaN, "len8_arg_accuracy": NaN, "len8_program_exact": NaN, "len12_direct_accuracy": 0.01171875, "len12_direct_target_mass": 0.012189237400889397, "len12_executor_accuracy": NaN, "len12_executor_target_mass": NaN, "len12_init_accuracy": NaN, "len12_op_accuracy": NaN, "len12_arg_accuracy": NaN, "len12_program_exact": NaN }, { "step": 200, "loss": 4.373783111572266, "direct_loss": 4.373783111572266, "len4_direct_accuracy": 0.00390625, "len4_direct_target_mass": 0.00990230031311512, "len4_executor_accuracy": NaN, "len4_executor_target_mass": NaN, "len4_init_accuracy": NaN, "len4_op_accuracy": NaN, "len4_arg_accuracy": NaN, "len4_program_exact": NaN, "len8_direct_accuracy": 0.01171875, "len8_direct_target_mass": 0.0112072778865695, "len8_executor_accuracy": NaN, "len8_executor_target_mass": NaN, "len8_init_accuracy": NaN, "len8_op_accuracy": NaN, "len8_arg_accuracy": NaN, "len8_program_exact": NaN, "len12_direct_accuracy": 0.0078125, "len12_direct_target_mass": 0.01029918622225523, "len12_executor_accuracy": NaN, "len12_executor_target_mass": NaN, "len12_init_accuracy": NaN, "len12_op_accuracy": NaN, "len12_arg_accuracy": NaN, "len12_program_exact": NaN }, { "step": 400, "loss": 4.198248386383057, "direct_loss": 4.198248386383057, "len4_direct_accuracy": 0.0078125, "len4_direct_target_mass": 0.010002022609114647, "len4_executor_accuracy": NaN, "len4_executor_target_mass": NaN, "len4_init_accuracy": NaN, "len4_op_accuracy": NaN, "len4_arg_accuracy": NaN, "len4_program_exact": NaN, "len8_direct_accuracy": 0.01171875, "len8_direct_target_mass": 0.011109495535492897, "len8_executor_accuracy": NaN, "len8_executor_target_mass": NaN, "len8_init_accuracy": NaN, "len8_op_accuracy": NaN, "len8_arg_accuracy": NaN, "len8_program_exact": NaN, "len12_direct_accuracy": 0.01171875, "len12_direct_target_mass": 0.009726714342832565, "len12_executor_accuracy": NaN, "len12_executor_target_mass": NaN, "len12_init_accuracy": NaN, "len12_op_accuracy": NaN, "len12_arg_accuracy": NaN, "len12_program_exact": NaN }, { "step": 600, "loss": 4.1198835372924805, "direct_loss": 4.1198835372924805, "len4_direct_accuracy": 0.0, "len4_direct_target_mass": 0.010062765330076218, "len4_executor_accuracy": NaN, "len4_executor_target_mass": NaN, "len4_init_accuracy": NaN, "len4_op_accuracy": NaN, "len4_arg_accuracy": NaN, "len4_program_exact": NaN, "len8_direct_accuracy": 0.015625, "len8_direct_target_mass": 0.011199703440070152, "len8_executor_accuracy": NaN, "len8_executor_target_mass": NaN, "len8_init_accuracy": NaN, "len8_op_accuracy": NaN, "len8_arg_accuracy": NaN, "len8_program_exact": NaN, "len12_direct_accuracy": 0.015625, "len12_direct_target_mass": 0.009607265703380108, "len12_executor_accuracy": NaN, "len12_executor_target_mass": NaN, "len12_init_accuracy": NaN, "len12_op_accuracy": NaN, "len12_arg_accuracy": NaN, "len12_program_exact": NaN }, { "step": 800, "loss": 3.9491491317749023, "direct_loss": 3.9491491317749023, "len4_direct_accuracy": 0.0, "len4_direct_target_mass": 0.009409862570464611, "len4_executor_accuracy": NaN, "len4_executor_target_mass": NaN, "len4_init_accuracy": NaN, "len4_op_accuracy": NaN, "len4_arg_accuracy": NaN, "len4_program_exact": NaN, "len8_direct_accuracy": 0.01171875, "len8_direct_target_mass": 0.011173621751368046, "len8_executor_accuracy": NaN, "len8_executor_target_mass": NaN, "len8_init_accuracy": NaN, "len8_op_accuracy": NaN, "len8_arg_accuracy": NaN, "len8_program_exact": NaN, "len12_direct_accuracy": 0.00390625, "len12_direct_target_mass": 0.009167246520519257, "len12_executor_accuracy": NaN, "len12_executor_target_mass": NaN, "len12_init_accuracy": NaN, "len12_op_accuracy": NaN, "len12_arg_accuracy": NaN, "len12_program_exact": NaN } ], "final_metrics": { "len4": { "variant": "direct", "n": 256.0, "direct_accuracy": 0.0, "direct_target_mass": 0.009409862570464611, "executor_accuracy": NaN, "executor_target_mass": NaN, "init_accuracy": NaN, "op_accuracy": NaN, "arg_accuracy": NaN, "program_exact": NaN }, "len8": { "variant": "direct", "n": 256.0, "direct_accuracy": 0.01171875, "direct_target_mass": 0.011173621751368046, "executor_accuracy": NaN, "executor_target_mass": NaN, "init_accuracy": NaN, "op_accuracy": NaN, "arg_accuracy": NaN, "program_exact": NaN }, "len12": { "variant": "direct", "n": 256.0, "direct_accuracy": 0.00390625, "direct_target_mass": 0.009167246520519257, "executor_accuracy": NaN, "executor_target_mass": NaN, "init_accuracy": NaN, "op_accuracy": NaN, "arg_accuracy": NaN, "program_exact": NaN } } }, "compiler_trace": { "variant": "compiler_trace", "train_seconds": 41.345704078674316, "checkpoints": [ "large_artifacts/qwen_structured_bridge/checkpoints/main_qwen35_numeric_spans/compiler_trace/checkpoint_final.pt" ], "train_log": [ { "step": 1, "loss": 14.832237243652344, "executor_loss": 4.567848205566406, "init_loss": 4.579281806945801, "op_loss": 1.0976622104644775, "arg_loss": 4.587445259094238, "len4_direct_accuracy": NaN, "len4_direct_target_mass": NaN, "len4_executor_accuracy": 0.0078125, "len4_executor_target_mass": 0.010314896702766418, "len4_init_accuracy": 0.03125, "len4_op_accuracy": 0.3193359375, "len4_arg_accuracy": 0.125, "len4_program_exact": 0.0, "len8_direct_accuracy": NaN, "len8_direct_target_mass": NaN, "len8_executor_accuracy": 0.015625, "len8_executor_target_mass": 0.010314551182091236, "len8_init_accuracy": 0.0078125, "len8_op_accuracy": 0.34130859375, "len8_arg_accuracy": 0.1103515625, "len8_program_exact": 0.0, "len12_direct_accuracy": NaN, "len12_direct_target_mass": NaN, "len12_executor_accuracy": 0.01171875, "len12_executor_target_mass": 0.010310172103345394, "len12_init_accuracy": 0.01953125, "len12_op_accuracy": 0.3470052083333333, "len12_arg_accuracy": 0.13704427083333334, "len12_program_exact": 0.0 }, { "step": 200, "loss": 0.0023916526697576046, "executor_loss": 0.0013837204314768314, "init_loss": 0.0007449277909472585, "op_loss": 9.748538286658004e-05, "arg_loss": 0.00016551915905438364, "len4_direct_accuracy": NaN, "len4_direct_target_mass": NaN, "len4_executor_accuracy": 1.0, "len4_executor_target_mass": 0.9944076538085938, "len4_init_accuracy": 1.0, "len4_op_accuracy": 1.0, "len4_arg_accuracy": 1.0, "len4_program_exact": 1.0, "len8_direct_accuracy": NaN, "len8_direct_target_mass": NaN, "len8_executor_accuracy": 0.984375, "len8_executor_target_mass": 0.9739116430282593, "len8_init_accuracy": 1.0, "len8_op_accuracy": 0.998046875, "len8_arg_accuracy": 1.0, "len8_program_exact": 0.984375, "len12_direct_accuracy": NaN, "len12_direct_target_mass": NaN, "len12_executor_accuracy": 0.93359375, "len12_executor_target_mass": 0.9159992337226868, "len12_init_accuracy": 1.0, "len12_op_accuracy": 0.9938151041666666, "len12_arg_accuracy": 1.0, "len12_program_exact": 0.93359375 }, { "step": 400, "loss": 0.0008322928333655, "executor_loss": 0.00046454803668893874, "init_loss": 0.0002968768822029233, "op_loss": 1.8182072381023318e-05, "arg_loss": 5.2685860282508656e-05, "len4_direct_accuracy": NaN, "len4_direct_target_mass": NaN, "len4_executor_accuracy": 1.0, "len4_executor_target_mass": 0.9968922734260559, "len4_init_accuracy": 1.0, "len4_op_accuracy": 1.0, "len4_arg_accuracy": 1.0, "len4_program_exact": 1.0, "len8_direct_accuracy": NaN, "len8_direct_target_mass": NaN, "len8_executor_accuracy": 0.9921875, "len8_executor_target_mass": 0.9840294122695923, "len8_init_accuracy": 1.0, "len8_op_accuracy": 0.9990234375, "len8_arg_accuracy": 1.0, "len8_program_exact": 0.9921875, "len12_direct_accuracy": NaN, "len12_direct_target_mass": NaN, "len12_executor_accuracy": 0.9453125, "len12_executor_target_mass": 0.9324361085891724, "len12_init_accuracy": 1.0, "len12_op_accuracy": 0.9954427083333334, "len12_arg_accuracy": 1.0, "len12_program_exact": 0.9453125 }, { "step": 600, "loss": 0.0003592233988456428, "executor_loss": 0.00020121067063882947, "init_loss": 0.0001275886024814099, "op_loss": 3.4219419831060804e-06, "arg_loss": 2.7002177375834435e-05, "len4_direct_accuracy": NaN, "len4_direct_target_mass": NaN, "len4_executor_accuracy": 1.0, "len4_executor_target_mass": 0.997199296951294, "len4_init_accuracy": 1.0, "len4_op_accuracy": 1.0, "len4_arg_accuracy": 1.0, "len4_program_exact": 1.0, "len8_direct_accuracy": NaN, "len8_direct_target_mass": NaN, "len8_executor_accuracy": 0.9921875, "len8_executor_target_mass": 0.9856027364730835, "len8_init_accuracy": 1.0, "len8_op_accuracy": 0.9990234375, "len8_arg_accuracy": 1.0, "len8_program_exact": 0.9921875, "len12_direct_accuracy": NaN, "len12_direct_target_mass": NaN, "len12_executor_accuracy": 0.9453125, "len12_executor_target_mass": 0.9351365566253662, "len12_init_accuracy": 1.0, "len12_op_accuracy": 0.9954427083333334, "len12_arg_accuracy": 1.0, "len12_program_exact": 0.9453125 }, { "step": 800, "loss": 0.00028427044162526727, "executor_loss": 0.000160602547111921, "init_loss": 0.0001001501950668171, "op_loss": 6.6455704654799774e-06, "arg_loss": 1.6872110791155137e-05, "len4_direct_accuracy": NaN, "len4_direct_target_mass": NaN, "len4_executor_accuracy": 1.0, "len4_executor_target_mass": 0.9980970621109009, "len4_init_accuracy": 1.0, "len4_op_accuracy": 1.0, "len4_arg_accuracy": 1.0, "len4_program_exact": 1.0, "len8_direct_accuracy": NaN, "len8_direct_target_mass": NaN, "len8_executor_accuracy": 0.9921875, "len8_executor_target_mass": 0.9883735179901123, "len8_init_accuracy": 1.0, "len8_op_accuracy": 0.9990234375, "len8_arg_accuracy": 1.0, "len8_program_exact": 0.9921875, "len12_direct_accuracy": NaN, "len12_direct_target_mass": NaN, "len12_executor_accuracy": 0.95703125, "len12_executor_target_mass": 0.9411150217056274, "len12_init_accuracy": 1.0, "len12_op_accuracy": 0.9964192708333334, "len12_arg_accuracy": 1.0, "len12_program_exact": 0.95703125 } ], "final_metrics": { "len4": { "variant": "compiler_trace", "n": 256.0, "direct_accuracy": NaN, "direct_target_mass": NaN, "executor_accuracy": 1.0, "executor_target_mass": 0.9980970621109009, "init_accuracy": 1.0, "op_accuracy": 1.0, "arg_accuracy": 1.0, "program_exact": 1.0 }, "len8": { "variant": "compiler_trace", "n": 256.0, "direct_accuracy": NaN, "direct_target_mass": NaN, "executor_accuracy": 0.9921875, "executor_target_mass": 0.9883735179901123, "init_accuracy": 1.0, "op_accuracy": 0.9990234375, "arg_accuracy": 1.0, "program_exact": 0.9921875 }, "len12": { "variant": "compiler_trace", "n": 256.0, "direct_accuracy": NaN, "direct_target_mass": NaN, "executor_accuracy": 0.95703125, "executor_target_mass": 0.9411150217056274, "init_accuracy": 1.0, "op_accuracy": 0.9964192708333334, "arg_accuracy": 1.0, "program_exact": 0.95703125 } } }, "compiler_answer_only": { "variant": "compiler_answer_only", "train_seconds": 41.79945993423462, "checkpoints": [ "large_artifacts/qwen_structured_bridge/checkpoints/main_qwen35_numeric_spans/compiler_answer_only/checkpoint_final.pt" ], "train_log": [ { "step": 1, "loss": 4.571533203125, "executor_loss": 4.571533203125, "len4_direct_accuracy": NaN, "len4_direct_target_mass": NaN, "len4_executor_accuracy": 0.01171875, "len4_executor_target_mass": 0.011732500046491623, "len4_init_accuracy": 0.0, "len4_op_accuracy": 0.35546875, "len4_arg_accuracy": 0.0, "len4_program_exact": 0.0, "len8_direct_accuracy": NaN, "len8_direct_target_mass": NaN, "len8_executor_accuracy": 0.01171875, "len8_executor_target_mass": 0.011552836745977402, "len8_init_accuracy": 0.01171875, "len8_op_accuracy": 0.3388671875, "len8_arg_accuracy": 0.0, "len8_program_exact": 0.0, "len12_direct_accuracy": NaN, "len12_direct_target_mass": NaN, "len12_executor_accuracy": 0.0078125, "len12_executor_target_mass": 0.007658686488866806, "len12_init_accuracy": 0.0078125, "len12_op_accuracy": 0.333984375, "len12_arg_accuracy": 0.0, "len12_program_exact": 0.0 }, { "step": 200, "loss": 4.37475061416626, "executor_loss": 4.37475061416626, "len4_direct_accuracy": NaN, "len4_direct_target_mass": NaN, "len4_executor_accuracy": 0.015625, "len4_executor_target_mass": 0.01031133346259594, "len4_init_accuracy": 0.0, "len4_op_accuracy": 0.310546875, "len4_arg_accuracy": 0.0166015625, "len4_program_exact": 0.0, "len8_direct_accuracy": NaN, "len8_direct_target_mass": NaN, "len8_executor_accuracy": 0.02734375, "len8_executor_target_mass": 0.010328639298677444, "len8_init_accuracy": 0.01171875, "len8_op_accuracy": 0.32861328125, "len8_arg_accuracy": 0.01953125, "len8_program_exact": 0.0, "len12_direct_accuracy": NaN, "len12_direct_target_mass": NaN, "len12_executor_accuracy": 0.01953125, "len12_executor_target_mass": 0.010309964418411255, "len12_init_accuracy": 0.01171875, "len12_op_accuracy": 0.3349609375, "len12_arg_accuracy": 0.024088541666666668, "len12_program_exact": 0.0 }, { "step": 400, "loss": 3.6648499965667725, "executor_loss": 3.6648499965667725, "len4_direct_accuracy": NaN, "len4_direct_target_mass": NaN, "len4_executor_accuracy": 0.0078125, "len4_executor_target_mass": 0.010479317046701908, "len4_init_accuracy": 0.0, "len4_op_accuracy": 0.310546875, "len4_arg_accuracy": 0.0185546875, "len4_program_exact": 0.0, "len8_direct_accuracy": NaN, "len8_direct_target_mass": NaN, "len8_executor_accuracy": 0.01171875, "len8_executor_target_mass": 0.010263081640005112, "len8_init_accuracy": 0.01171875, "len8_op_accuracy": 0.32861328125, "len8_arg_accuracy": 0.0185546875, "len8_program_exact": 0.0, "len12_direct_accuracy": NaN, "len12_direct_target_mass": NaN, "len12_executor_accuracy": 0.0078125, "len12_executor_target_mass": 0.010302460752427578, "len12_init_accuracy": 0.01171875, "len12_op_accuracy": 0.3349609375, "len12_arg_accuracy": 0.027018229166666668, "len12_program_exact": 0.0 }, { "step": 600, "loss": 2.919240951538086, "executor_loss": 2.919240951538086, "len4_direct_accuracy": NaN, "len4_direct_target_mass": NaN, "len4_executor_accuracy": 0.0078125, "len4_executor_target_mass": 0.00895591452717781, "len4_init_accuracy": 0.0, "len4_op_accuracy": 0.310546875, "len4_arg_accuracy": 0.03125, "len4_program_exact": 0.0, "len8_direct_accuracy": NaN, "len8_direct_target_mass": NaN, "len8_executor_accuracy": 0.0, "len8_executor_target_mass": 0.010112661868333817, "len8_init_accuracy": 0.01171875, "len8_op_accuracy": 0.32861328125, "len8_arg_accuracy": 0.03173828125, "len8_program_exact": 0.0, "len12_direct_accuracy": NaN, "len12_direct_target_mass": NaN, "len12_executor_accuracy": 0.00390625, "len12_executor_target_mass": 0.010286466218531132, "len12_init_accuracy": 0.01171875, "len12_op_accuracy": 0.3349609375, "len12_arg_accuracy": 0.040690104166666664, "len12_program_exact": 0.0 }, { "step": 800, "loss": 2.5908682346343994, "executor_loss": 2.5908682346343994, "len4_direct_accuracy": NaN, "len4_direct_target_mass": NaN, "len4_executor_accuracy": 0.015625, "len4_executor_target_mass": 0.009302075020968914, "len4_init_accuracy": 0.0, "len4_op_accuracy": 0.310546875, "len4_arg_accuracy": 0.0263671875, "len4_program_exact": 0.0, "len8_direct_accuracy": NaN, "len8_direct_target_mass": NaN, "len8_executor_accuracy": 0.0, "len8_executor_target_mass": 0.009441904723644257, "len8_init_accuracy": 0.01171875, "len8_op_accuracy": 0.32861328125, "len8_arg_accuracy": 0.0244140625, "len8_program_exact": 0.0, "len12_direct_accuracy": NaN, "len12_direct_target_mass": NaN, "len12_executor_accuracy": 0.0078125, "len12_executor_target_mass": 0.010155543684959412, "len12_init_accuracy": 0.01171875, "len12_op_accuracy": 0.3349609375, "len12_arg_accuracy": 0.030598958333333332, "len12_program_exact": 0.0 } ], "final_metrics": { "len4": { "variant": "compiler_answer_only", "n": 256.0, "direct_accuracy": NaN, "direct_target_mass": NaN, "executor_accuracy": 0.015625, "executor_target_mass": 0.009302075020968914, "init_accuracy": 0.0, "op_accuracy": 0.310546875, "arg_accuracy": 0.0263671875, "program_exact": 0.0 }, "len8": { "variant": "compiler_answer_only", "n": 256.0, "direct_accuracy": NaN, "direct_target_mass": NaN, "executor_accuracy": 0.0, "executor_target_mass": 0.009441904723644257, "init_accuracy": 0.01171875, "op_accuracy": 0.32861328125, "arg_accuracy": 0.0244140625, "program_exact": 0.0 }, "len12": { "variant": "compiler_answer_only", "n": 256.0, "direct_accuracy": NaN, "direct_target_mass": NaN, "executor_accuracy": 0.0078125, "executor_target_mass": 0.010155543684959412, "init_accuracy": 0.01171875, "op_accuracy": 0.3349609375, "arg_accuracy": 0.030598958333333332, "program_exact": 0.0 } } } } }