{ "schema": "public_first_common30_test_and_remaining_budget_v1", "as_of_beijing": "2026-11-02 16:37:20", "budget_anchor_utc": "2026-11-02T08:36:10.340434+00:00", "training": { "completed_main_rounds": 21, "total_main_rounds": 21, "all_main_training_complete": true, "auxiliary_additional_updates_still_planned": 3, "final_common_test_complete": false }, "completed_round": { "arm": "WDT_SCALE50", "round": 3, "status": "COMPLETE", "seed_questions": 50, "review_questions": 25, "new_questions": 9, "questions": 84, "answers": 672, "answers_per_question": 8, "effective_questions": 40, "effective_new_questions": 4, "all_succeeded_questions": 33, "all_failed_questions": 11, "new_main_learning_updates": 1, "final_model_saved": true, "repair_enabled": false, "new_question_breakdown": { "deep_down": { "all_failed": 0, "all_succeeded": 1, "effective_questions": 0, "minimum_actions_delta_range": [ -10, -10 ], "questions": 1, "required_chain_depth_delta_range": [ -1, -1 ] }, "deep_up": { "all_failed": 0, "all_succeeded": 0, "effective_questions": 2, "minimum_actions_delta_range": [ 1, 4 ], "questions": 2, "required_chain_depth_delta_range": [ 1, 1 ] }, "wide": { "all_failed": 1, "all_succeeded": 3, "effective_questions": 2, "minimum_actions_delta_range": [ -3, 1 ], "questions": 6, "required_chain_depth_delta_range": [ -1, 1 ] } }, "seed_observations": { "questions": 50, "previous_round": 2, "current_round": 3, "previous_successes": 279, "previous_answers": 400, "current_successes": 279, "current_answers": 400, "improved_questions": 12, "regressed_questions": 17, "unchanged_questions": 21, "recovered_from_all_failed": 0, "regressed_to_all_failed": 2 }, "common_old_task_observations": { "questions": 75, "previous_round": 2, "current_round": 3, "previous_successes": 402, "previous_answers": 600, "current_successes": 407, "current_answers": 600, "improved_questions": 20, "regressed_questions": 22, "unchanged_questions": 33, "recovered_from_all_failed": 1, "regressed_to_all_failed": 4 }, "observation_interpretation": "第二轮与第三轮作答的比较；第三轮答卷来自第二轮最终模型。第三轮训练后的最终表现由共同测试衡量。", "start_utc": "2026-11-02T06:22:35.695521+00:00", "end_utc": "2026-11-02T08:10:21.268705+00:00", "start_beijing": "2026-11-02 14:22:35", "end_beijing": "2026-11-02 16:10:21", "full_elapsed_seconds": 6465.573184, "stage_sum_seconds": 6465.546611000001, "interstage_seconds": 0.026573, "recorded_stage_seconds": { "plan-round": 4874.832532, "train-round": 1590.714079 }, "pure_parameter_training_seconds": null, "elapsed_scope": "整轮用时包含出题、作答、学习、保存及环节交接；后半段也不是纯模型调整时间。" }, "remaining_work": { "main_training_rounds_completed": 21, "main_training_rounds_total": 21, "remaining_main_training_rounds": 0, "main_total_models": 8, "main_completed_models_at_anchor": 1, "main_remaining_models_at_anchor": 7, "main_total_test_answers": 720, "main_completed_full_model_test_answers": 90, "main_remaining_test_answers_at_anchor": 630, "auxiliary_test_models": 2, "auxiliary_final_test_answers": 180, "auxiliary_additional_updates": 3, "remaining_test_answers_at_anchor": 810, "final_return_and_stop_included": true, "later_WT_partial_answers_not_deducted": 4 }, "budget_summary": { "as_of_beijing": "2026-11-02 16:36:10", "central_remaining": "约 4 小时 30 分", "central_complete": "2026-11-02 21:03 左右", "planning_range": "约 2 小时 20 分–11 小时 50 分", "faster_complete": "2026-11-02 18:56 左右", "slower_complete": "2026-11-03 04:18 左右", "gpu_cost_central_usd": 8, "gpu_cost_range_usd": [ 4, 21 ], "scope": "剩余 7 个主要测试模型共 630 份回答，2 个小对照模型共 180 份测试，追加 3 次学习和收尾。", "rate_note": "沿用每小时 1.79 美元的已有记录，只计后续显卡费用。" }, "scenarios": [ { "case": "faster", "remaining_seconds": 8394.15546311047, "remaining_hours": 2.3317098508640193, "display_duration": "约 2 小时 20 分", "estimated_complete_beijing_calculated": "2026-11-02T18:56:04.495897+08:00", "estimated_complete_beijing_display": "2026-11-02 18:56 左右", "gpu_usd_at_recorded_rate": 4.173760633046594, "display_gpu_usd": 4, "components_seconds": { "seven_remaining_model_common30_test_630_answers": 5383.1031942336, "fixed_batch_additional_three_updates_and_initial_probabilities": 1173.0227848101267, "fixed_batch_common30_test_180_answers": 1538.029484066743, "final_return_and_stop_allowance": 300 }, "test_factor_relative_to_central": 0.49645714285714293 }, { "case": "central", "remaining_seconds": 16007.326118012657, "remaining_hours": 4.446479477225738, "display_duration": "约 4 小时 30 分", "estimated_complete_beijing_calculated": "2026-11-02T21:02:57.666552+08:00", "estimated_complete_beijing_display": "2026-11-02 21:03 左右", "gpu_usd_at_recorded_rate": 7.959198264234072, "display_gpu_usd": 8, "components_seconds": { "seven_remaining_model_common30_test_630_answers": 10843.037051, "fixed_batch_additional_three_updates_and_initial_probabilities": 1466.2784810126584, "fixed_batch_common30_test_180_answers": 3098.010586, "final_return_and_stop_allowance": 600 }, "test_factor_relative_to_central": 1.0 }, { "case": "slower", "remaining_seconds": 42080.12988572279, "remaining_hours": 11.68892496825633, "display_duration": "约 11 小时 50 分", "estimated_complete_beijing_calculated": "2026-11-03T04:17:30.470320+08:00", "estimated_complete_beijing_display": "2026-11-03 04:18 左右", "gpu_usd_at_recorded_rate": 20.92317569317883, "display_gpu_usd": 21, "components_seconds": { "seven_remaining_model_common30_test_630_answers": 29276.2000377, "fixed_batch_additional_three_updates_and_initial_probabilities": 2639.3012658227854, "fixed_batch_common30_test_180_answers": 8364.6285822, "final_return_and_stop_allowance": 1800 }, "test_factor_relative_to_central": 2.7 } ], "budget_basis": "N 一组完整测试实测约 25 分 49 秒。余下 7 组主测试和 2 组小对照测试暂按相同用时估计；追加学习和收尾保留原额度。WT 的 4 份已存回答没有另扣。", "cost": { "gpu_hourly_usd_recorded": 1.79, "rate_observed_utc": "2026-11-01T14:07:02.455461+00:00", "rate_rechecked": false, "teacher_charges_included": false, "storage_charges_included": false, "current_accumulated_bill_known": false }, "budget_anchor_beijing": "2026-11-02 16:36:10", "completed_test": { "label": "N", "complete": true, "summary_matches_scope": true, "questions": 30, "answers": 90, "sampled_successes": 27, "sampled_attempts": 60, "sampled_rate": 0.45, "greedy_successes": 15, "greedy_attempts": 30, "greedy_rate": 0.5 }, "latest_test_observation": { "as_of_utc": "2026-11-02T08:37:20.782629+00:00", "as_of_beijing": "2026-11-02 16:37:20", "main_completed_models": 1, "main_required_models": 8, "main_answers_saved": 94, "main_answers_required": 720, "WT_answers_saved": 4, "WT_complete": false, "auxiliary_answers_saved": 0, "auxiliary_answers_required": 180, "partial_WT_answers_deducted_from_budget": false }, "measured_test_interval": { "model": "N", "next_model": "WT", "start_utc": "2026-11-02T08:10:21.335141+00:00", "next_stage_start_utc": "2026-11-02T08:36:10.340434+00:00", "full_interval_seconds": 1549.005293, "full_interval_minutes": 25.81675488333333, "scope": "从 N 测试开始到 WT 测试开始，包含加载、测试、收尾和交接。", "pure_answer_generation_seconds": null, "observed_complete_model_intervals": 1, "assumed_remaining_main_models": 7, "assumed_auxiliary_models": 2, "remaining_models_have_measured_test_speed": false, "prior_central_single_model_seconds": 1333.9522722720994, "ratio_to_prior_central_single_model": 1.1612149288981712, "basis": "原有观察记录中的阶段开始时间；没有发起新观察。" }, "scenario_basis": "沿用前一预算各测试项相对中间情景的倍率；小对照学习和收尾保留各情景原额度。", "source_records": { "last_training_round": { "record_id": "SCALE50_R3_COMPLETED_20261102_CAPTURE01", "capture_sha256": "af87ede171f84c1fcba3e26909a05feecb355986ea23b12819c1cffe12dcb4b9" }, "N_test_start": { "record": "local_observer_20261101T220625Z.jsonl", "line": 807, "line_sha256": "07684bef82fac338ae7e7994aa8379479bb2f8fd196a4c7deaf6c151c91d587b", "observation_utc": "2026-11-02T08:11:06.548393+00:00" }, "N_test_complete_and_WT_start": { "record": "local_observer_20261101T220625Z.jsonl", "line": 842, "line_sha256": "1da9ed6f162bd8da3f6b44c98b5a1f426d0fa51d3184dd261fe9f42d80935f0d", "observation_utc": "2026-11-02T08:37:20.782629+00:00" } } }