{ "date": "2026-09-29", "code_revision": "d049fd293e3232358594cba7cdae0f7a138bb58d", "base_release": "d39842eeca8132288c5e2814fdd66733fa92fc3d", "bundle": "personamem-32k/qwen3-4b", "phase": "on_policy_optimization", "selection_criterion": "Highest full-pipeline validation mean over five deterministic option permutations; tie: earlier epoch then configuration order. Evaluate test only after selection.", "selected_configuration": { "name": "more-grpo", "learning_rate": 1.5e-07, "opd_weight": 0.0010033222591362126, "grpo_weight": 0.30099667774086375 }, "selected_epoch": 5, "selected_optimizer_steps": 615, "reference_kl_weight": 0, "teacher_update_mode": "frozen_initial", "epochs_per_configuration": 5, "world_size": 4, "questions_per_update": 4, "rollouts_per_question": 8, "rollouts_per_update": 32, "seed": 42, "sampling_temperature": 1.0, "top_p": 0.98, "max_response_tokens": 5, "training_examples": 489, "validation_examples": 50, "test_examples": 50, "evaluation": "Full-pipeline greedy memory generation and five fixed option permutations; test evaluated after validation selection.", "validation_delta_vs_incumbent": 0.02400000000000002, "validation_results": [ { "name": "incumbent", "trials": [ { "trial": 0, "correct": 34, "total": 50, "accuracy": 0.68 }, { "trial": 1, "correct": 33, "total": 50, "accuracy": 0.66 }, { "trial": 2, "correct": 38, "total": 50, "accuracy": 0.76 }, { "trial": 3, "correct": 39, "total": 50, "accuracy": 0.78 }, { "trial": 4, "correct": 38, "total": 50, "accuracy": 0.76 } ], "mean_accuracy": 0.728, "best_accuracy": 0.78, "seconds": 38.45063753426075 }, { "name": "current-ratio-epoch-1", "trials": [ { "trial": 0, "correct": 33, "total": 50, "accuracy": 0.66 }, { "trial": 1, "correct": 32, "total": 50, "accuracy": 0.64 }, { "trial": 2, "correct": 40, "total": 50, "accuracy": 0.8 }, { "trial": 3, "correct": 39, "total": 50, "accuracy": 0.78 }, { "trial": 4, "correct": 34, "total": 50, "accuracy": 0.68 } ], "mean_accuracy": 0.712, "best_accuracy": 0.8, "seconds": 42.41846241056919 }, { "name": "current-ratio-epoch-2", "trials": [ { "trial": 0, "correct": 35, "total": 50, "accuracy": 0.7 }, { "trial": 1, "correct": 33, "total": 50, "accuracy": 0.66 }, { "trial": 2, "correct": 38, "total": 50, "accuracy": 0.76 }, { "trial": 3, "correct": 41, "total": 50, "accuracy": 0.82 }, { "trial": 4, "correct": 36, "total": 50, "accuracy": 0.72 } ], "mean_accuracy": 0.732, "best_accuracy": 0.82, "seconds": 41.58905404154211 }, { "name": "current-ratio-epoch-3", "trials": [ { "trial": 0, "correct": 35, "total": 50, "accuracy": 0.7 }, { "trial": 1, "correct": 33, "total": 50, "accuracy": 0.66 }, { "trial": 2, "correct": 40, "total": 50, "accuracy": 0.8 }, { "trial": 3, "correct": 41, "total": 50, "accuracy": 0.82 }, { "trial": 4, "correct": 37, "total": 50, "accuracy": 0.74 } ], "mean_accuracy": 0.744, "best_accuracy": 0.82, "seconds": 43.19170383922756 }, { "name": "current-ratio-epoch-4", "trials": [ { "trial": 0, "correct": 35, "total": 50, "accuracy": 0.7 }, { "trial": 1, "correct": 31, "total": 50, "accuracy": 0.62 }, { "trial": 2, "correct": 40, "total": 50, "accuracy": 0.8 }, { "trial": 3, "correct": 37, "total": 50, "accuracy": 0.74 }, { "trial": 4, "correct": 37, "total": 50, "accuracy": 0.74 } ], "mean_accuracy": 0.7200000000000001, "best_accuracy": 0.8, "seconds": 42.76385089289397 }, { "name": "current-ratio-epoch-5", "trials": [ { "trial": 0, "correct": 36, "total": 50, "accuracy": 0.72 }, { "trial": 1, "correct": 34, "total": 50, "accuracy": 0.68 }, { "trial": 2, "correct": 39, "total": 50, "accuracy": 0.78 }, { "trial": 3, "correct": 38, "total": 50, "accuracy": 0.76 }, { "trial": 4, "correct": 38, "total": 50, "accuracy": 0.76 } ], "mean_accuracy": 0.7399999999999999, "best_accuracy": 0.78, "seconds": 51.30520835891366 }, { "name": "lower-lr-epoch-1", "trials": [ { "trial": 0, "correct": 35, "total": 50, "accuracy": 0.7 }, { "trial": 1, "correct": 33, "total": 50, "accuracy": 0.66 }, { "trial": 2, "correct": 37, "total": 50, "accuracy": 0.74 }, { "trial": 3, "correct": 39, "total": 50, "accuracy": 0.78 }, { "trial": 4, "correct": 35, "total": 50, "accuracy": 0.7 } ], "mean_accuracy": 0.716, "best_accuracy": 0.78, "seconds": 50.440840646624565 }, { "name": "lower-lr-epoch-2", "trials": [ { "trial": 0, "correct": 34, "total": 50, "accuracy": 0.68 }, { "trial": 1, "correct": 32, "total": 50, "accuracy": 0.64 }, { "trial": 2, "correct": 38, "total": 50, "accuracy": 0.76 }, { "trial": 3, "correct": 39, "total": 50, "accuracy": 0.78 }, { "trial": 4, "correct": 37, "total": 50, "accuracy": 0.74 } ], "mean_accuracy": 0.7200000000000001, "best_accuracy": 0.78, "seconds": 43.54510813206434 }, { "name": "lower-lr-epoch-3", "trials": [ { "trial": 0, "correct": 36, "total": 50, "accuracy": 0.72 }, { "trial": 1, "correct": 32, "total": 50, "accuracy": 0.64 }, { "trial": 2, "correct": 40, "total": 50, "accuracy": 0.8 }, { "trial": 3, "correct": 39, "total": 50, "accuracy": 0.78 }, { "trial": 4, "correct": 35, "total": 50, "accuracy": 0.7 } ], "mean_accuracy": 0.7280000000000001, "best_accuracy": 0.8, "seconds": 40.46544888615608 }, { "name": "lower-lr-epoch-4", "trials": [ { "trial": 0, "correct": 36, "total": 50, "accuracy": 0.72 }, { "trial": 1, "correct": 32, "total": 50, "accuracy": 0.64 }, { "trial": 2, "correct": 38, "total": 50, "accuracy": 0.76 }, { "trial": 3, "correct": 42, "total": 50, "accuracy": 0.84 }, { "trial": 4, "correct": 36, "total": 50, "accuracy": 0.72 } ], "mean_accuracy": 0.736, "best_accuracy": 0.84, "seconds": 48.450808545574546 }, { "name": "lower-lr-epoch-5", "trials": [ { "trial": 0, "correct": 35, "total": 50, "accuracy": 0.7 }, { "trial": 1, "correct": 33, "total": 50, "accuracy": 0.66 }, { "trial": 2, "correct": 37, "total": 50, "accuracy": 0.74 }, { "trial": 3, "correct": 42, "total": 50, "accuracy": 0.84 }, { "trial": 4, "correct": 38, "total": 50, "accuracy": 0.76 } ], "mean_accuracy": 0.7399999999999999, "best_accuracy": 0.84, "seconds": 38.6599240321666 }, { "name": "more-grpo-epoch-1", "trials": [ { "trial": 0, "correct": 33, "total": 50, "accuracy": 0.66 }, { "trial": 1, "correct": 32, "total": 50, "accuracy": 0.64 }, { "trial": 2, "correct": 38, "total": 50, "accuracy": 0.76 }, { "trial": 3, "correct": 38, "total": 50, "accuracy": 0.76 }, { "trial": 4, "correct": 34, "total": 50, "accuracy": 0.68 } ], "mean_accuracy": 0.7000000000000001, "best_accuracy": 0.76, "seconds": 43.131004037335515 }, { "name": "more-grpo-epoch-2", "trials": [ { "trial": 0, "correct": 35, "total": 50, "accuracy": 0.7 }, { "trial": 1, "correct": 34, "total": 50, "accuracy": 0.68 }, { "trial": 2, "correct": 39, "total": 50, "accuracy": 0.78 }, { "trial": 3, "correct": 42, "total": 50, "accuracy": 0.84 }, { "trial": 4, "correct": 33, "total": 50, "accuracy": 0.66 } ], "mean_accuracy": 0.732, "best_accuracy": 0.84, "seconds": 39.89130744803697 }, { "name": "more-grpo-epoch-3", "trials": [ { "trial": 0, "correct": 37, "total": 50, "accuracy": 0.74 }, { "trial": 1, "correct": 34, "total": 50, "accuracy": 0.68 }, { "trial": 2, "correct": 39, "total": 50, "accuracy": 0.78 }, { "trial": 3, "correct": 39, "total": 50, "accuracy": 0.78 }, { "trial": 4, "correct": 35, "total": 50, "accuracy": 0.7 } ], "mean_accuracy": 0.7360000000000001, "best_accuracy": 0.78, "seconds": 46.916815302334726 }, { "name": "more-grpo-epoch-4", "trials": [ { "trial": 0, "correct": 37, "total": 50, "accuracy": 0.74 }, { "trial": 1, "correct": 35, "total": 50, "accuracy": 0.7 }, { "trial": 2, "correct": 38, "total": 50, "accuracy": 0.76 }, { "trial": 3, "correct": 41, "total": 50, "accuracy": 0.82 }, { "trial": 4, "correct": 36, "total": 50, "accuracy": 0.72 } ], "mean_accuracy": 0.748, "best_accuracy": 0.82, "seconds": 42.778895856812596 }, { "name": "more-grpo-epoch-5", "trials": [ { "trial": 0, "correct": 37, "total": 50, "accuracy": 0.74 }, { "trial": 1, "correct": 34, "total": 50, "accuracy": 0.68 }, { "trial": 2, "correct": 38, "total": 50, "accuracy": 0.76 }, { "trial": 3, "correct": 43, "total": 50, "accuracy": 0.86 }, { "trial": 4, "correct": 36, "total": 50, "accuracy": 0.72 } ], "mean_accuracy": 0.752, "best_accuracy": 0.86, "seconds": 43.428868914954364 }, { "name": "more-opd-epoch-1", "trials": [ { "trial": 0, "correct": 34, "total": 50, "accuracy": 0.68 }, { "trial": 1, "correct": 33, "total": 50, "accuracy": 0.66 }, { "trial": 2, "correct": 40, "total": 50, "accuracy": 0.8 }, { "trial": 3, "correct": 42, "total": 50, "accuracy": 0.84 }, { "trial": 4, "correct": 36, "total": 50, "accuracy": 0.72 } ], "mean_accuracy": 0.74, "best_accuracy": 0.84, "seconds": 51.145253921858966 }, { "name": "more-opd-epoch-2", "trials": [ { "trial": 0, "correct": 35, "total": 50, "accuracy": 0.7 }, { "trial": 1, "correct": 30, "total": 50, "accuracy": 0.6 }, { "trial": 2, "correct": 38, "total": 50, "accuracy": 0.76 }, { "trial": 3, "correct": 41, "total": 50, "accuracy": 0.82 }, { "trial": 4, "correct": 36, "total": 50, "accuracy": 0.72 } ], "mean_accuracy": 0.72, "best_accuracy": 0.82, "seconds": 39.484272602945566 }, { "name": "more-opd-epoch-3", "trials": [ { "trial": 0, "correct": 36, "total": 50, "accuracy": 0.72 }, { "trial": 1, "correct": 33, "total": 50, "accuracy": 0.66 }, { "trial": 2, "correct": 39, "total": 50, "accuracy": 0.78 }, { "trial": 3, "correct": 42, "total": 50, "accuracy": 0.84 }, { "trial": 4, "correct": 37, "total": 50, "accuracy": 0.74 } ], "mean_accuracy": 0.748, "best_accuracy": 0.84, "seconds": 53.43376477062702 }, { "name": "more-opd-epoch-4", "trials": [ { "trial": 0, "correct": 35, "total": 50, "accuracy": 0.7 }, { "trial": 1, "correct": 32, "total": 50, "accuracy": 0.64 }, { "trial": 2, "correct": 38, "total": 50, "accuracy": 0.76 }, { "trial": 3, "correct": 41, "total": 50, "accuracy": 0.82 }, { "trial": 4, "correct": 37, "total": 50, "accuracy": 0.74 } ], "mean_accuracy": 0.7319999999999999, "best_accuracy": 0.82, "seconds": 47.37319718301296 }, { "name": "more-opd-epoch-5", "trials": [ { "trial": 0, "correct": 37, "total": 50, "accuracy": 0.74 }, { "trial": 1, "correct": 34, "total": 50, "accuracy": 0.68 }, { "trial": 2, "correct": 37, "total": 50, "accuracy": 0.74 }, { "trial": 3, "correct": 41, "total": 50, "accuracy": 0.82 }, { "trial": 4, "correct": 38, "total": 50, "accuracy": 0.76 } ], "mean_accuracy": 0.748, "best_accuracy": 0.82, "seconds": 38.24769929051399 } ], "selected_test": { "name": "selected", "trials": [ { "trial": 0, "correct": 41, "total": 50, "accuracy": 0.82 }, { "trial": 1, "correct": 38, "total": 50, "accuracy": 0.76 }, { "trial": 2, "correct": 34, "total": 50, "accuracy": 0.68 }, { "trial": 3, "correct": 39, "total": 50, "accuracy": 0.78 }, { "trial": 4, "correct": 38, "total": 50, "accuracy": 0.76 } ], "mean_accuracy": 0.76, "best_accuracy": 0.82, "seconds": 50.37944988813251 }, "train_seconds": { "current-ratio": 1361.2398540973663, "lower-lr": 1376.2528283596039, "more-grpo": 1360.094055362977, "more-opd": 1365.79950528685 }, "scope": "Only the final mixed-loss phase was retrained. Compressor and initialization held fixed. The candidate is not a replacement for historical paper results." }