{ "acceptance": { "accepted_for_mixed_serving_thermal_constraints": false, "all_criteria_passed": false, "criteria": { "asymptotically_stable": { "passed": true, "required": true, "value": true }, "full_jacobian_rank": { "jacobian_numerical_rank": 4, "parameter_count": 4, "passed": true, "required": true }, "locally_identifiable": { "condition_number_limit": 100000000.0, "passed": true, "required": true, "value": true }, "optimizer_success": { "passed": true, "required": true, "value": true }, "physical_parameters": { "passed": true, "required": true, "value": true }, "replicated_multistart_solution": { "minimum_required_for_each": 2, "near_optimal_multistarts_within_one_percent": 10, "passed": true, "required": true, "successful_multistarts": 10 }, "validation_maximum_absolute_or_peak_error": { "maximum_absolute_per_pulse_peak_error_c": 1.1894546061778755, "maximum_absolute_trajectory_error_c": 2.7310039794374816, "operator": "<", "passed": false, "threshold_c": 1.0, "value_c": 2.7310039794374816 }, "validation_phase_contrast_error": { "absolute_error_c": 1.431289590989124, "contrast_definition": "prefill minus decode peak temperature rise on the separate one-second validation grids", "observed_contrast_c": 2.0, "operator": "<", "passed": false, "predicted_contrast_c": 3.431289590989124, "signed_error_c": 1.431289590989124, "threshold_c": 1.0 }, "validation_rmse": { "operator": "<", "passed": true, "threshold_c": 1.0, "value_c": 0.8696553321716621 } }, "rule": "all validation errors are strictly below 1 C and the fitted parameters are physical, stable, locally identified, full-rank, and reproduced by at least two successful near-optimal starts", "target_model": "phase_gain_one_state_rc", "verdict": "rejected" }, "measured_power_and_energy": { "matched_phase_pairs": { "phase_training": { "46": { "decode_block_id": "phase_training_pulse_00", "prefill_block_id": "phase_training_pulse_02", "prefill_minus_decode_energy_j": -32.1978992984109, "prefill_minus_decode_mean_power_w": -0.4001689470273746, "prefill_to_decode_energy_ratio": 0.9907484796043877 }, "61": { "decode_block_id": "phase_training_pulse_03", "prefill_block_id": "phase_training_pulse_01", "prefill_minus_decode_energy_j": 12.085332144686163, "prefill_minus_decode_mean_power_w": 0.21912438525099986, "prefill_to_decode_energy_ratio": 1.0044500143189368 } }, "phase_validation": { "55": { "decode_block_id": "phase_validation_pulse_00", "prefill_block_id": "phase_validation_pulse_01", "prefill_minus_decode_energy_j": -6.664176432056593, "prefill_minus_decode_mean_power_w": -0.13225922291921677, "prefill_to_decode_energy_ratio": 0.9979678296957712 } } }, "per_pulse": { "phase_training_pulse_00": { "completed_batch_count": 20, "completed_request_rows": 160, "duration_s": 74.92543761399997, "end_temperature_c": 64.0, "first_peak_time_s": 69.52500172099997, "first_request_start_offset_s": -0.051125045000048885, "integrated_energy_j": 3480.2819343814344, "last_peak_time_s": 73.42413511799998, "last_request_end_offset_s": 1.5581068950000372, "maximum_realized_graphics_clock_mhz": 2040.0, "maximum_sampling_gap_s": 0.11039053500007867, "mean_utilization_percent": 98.90921228304406, "median_memory_clock_mhz": 6251.0, "median_realized_graphics_clock_mhz": 210.0, "minimum_realized_graphics_clock_mhz": 210.0, "peak_rise_c": 12.0, "peak_sample_count": 6, "peak_temperature_c": 65.0, "raw_rows": 749, "request_coverage_fraction": 0.9995598104722484, "requested_power_limit_w": 46.0, "start_temperature_c": 53.0, "terminal_rise_c": 11.0, "time_weighted_mean_power_w": 46.44993803454458, "workload_phase": "decode" }, "phase_training_pulse_01": { "completed_batch_count": 5, "completed_request_rows": 40, "duration_s": 44.906093021000174, "end_temperature_c": 70.0, "first_peak_time_s": 41.59651763299996, "first_request_start_offset_s": -0.05691981199993279, "integrated_energy_j": 2727.8815697562095, "last_peak_time_s": 44.906093021000174, "last_request_end_offset_s": 7.550731947999793, "maximum_realized_graphics_clock_mhz": 2040.0, "maximum_sampling_gap_s": 0.1082954940000036, "mean_utilization_percent": 99.47884187082406, "median_memory_clock_mhz": 6251.0, "median_realized_graphics_clock_mhz": 855.0, "minimum_realized_graphics_clock_mhz": 705.0, "peak_rise_c": 16.0, "peak_sample_count": 30, "peak_temperature_c": 70.0, "raw_rows": 449, "request_coverage_fraction": 0.9998008011519526, "requested_power_limit_w": 61.0, "start_temperature_c": 54.0, "terminal_rise_c": 16.0, "time_weighted_mean_power_w": 60.746357258925315, "workload_phase": "prefill" }, "phase_training_pulse_02": { "completed_batch_count": 5, "completed_request_rows": 40, "duration_s": 74.87733605199992, "end_temperature_c": 66.0, "first_peak_time_s": 73.16419886999984, "first_request_start_offset_s": -0.07496849100016334, "integrated_energy_j": 3448.0840350830235, "last_peak_time_s": 74.67715177399987, "last_request_end_offset_s": 14.953618433999964, "maximum_realized_graphics_clock_mhz": 780.0, "maximum_sampling_gap_s": 0.11270664300036515, "mean_utilization_percent": 99.68005354752343, "median_memory_clock_mhz": 6251.0, "median_realized_graphics_clock_mhz": 480.0, "minimum_realized_graphics_clock_mhz": 360.0, "peak_rise_c": 10.0, "peak_sample_count": 6, "peak_temperature_c": 67.0, "raw_rows": 747, "request_coverage_fraction": 0.9997704880554521, "requested_power_limit_w": 46.0, "start_temperature_c": 57.0, "terminal_rise_c": 9.0, "time_weighted_mean_power_w": 46.049769087517205, "workload_phase": "prefill" }, "phase_training_pulse_03": { "completed_batch_count": 21, "completed_request_rows": 168, "duration_s": 44.86899712200011, "end_temperature_c": 69.0, "first_peak_time_s": 44.86899712200011, "first_request_start_offset_s": -0.06943937899995944, "integrated_energy_j": 2715.7962376115233, "last_peak_time_s": 44.86899712200011, "last_request_end_offset_s": 1.9331424859997242, "maximum_realized_graphics_clock_mhz": 1725.0, "maximum_sampling_gap_s": 0.11982493300001806, "mean_utilization_percent": 98.71651785714286, "median_memory_clock_mhz": 6251.0, "median_realized_graphics_clock_mhz": 1140.0, "minimum_realized_graphics_clock_mhz": 615.0, "peak_rise_c": 13.0, "peak_sample_count": 1, "peak_temperature_c": 69.0, "raw_rows": 448, "request_coverage_fraction": 0.9991974290421181, "requested_power_limit_w": 61.0, "start_temperature_c": 56.0, "terminal_rise_c": 13.0, "time_weighted_mean_power_w": 60.527232873674315, "workload_phase": "decode" }, "phase_validation_pulse_00": { "completed_batch_count": 24, "completed_request_rows": 192, "duration_s": 59.92533942399996, "end_temperature_c": 68.0, "first_peak_time_s": 57.09476940700006, "first_request_start_offset_s": -0.06804972999998427, "integrated_energy_j": 3279.3395406816717, "last_peak_time_s": 59.92533942399996, "last_request_end_offset_s": 2.1620938320002097, "maximum_realized_graphics_clock_mhz": 1320.0, "maximum_sampling_gap_s": 0.1218580810000276, "mean_utilization_percent": 98.90050590219224, "median_memory_clock_mhz": 6251.0, "median_realized_graphics_clock_mhz": 345.0, "minimum_realized_graphics_clock_mhz": 270.0, "peak_rise_c": 12.0, "peak_sample_count": 5, "peak_temperature_c": 68.0, "raw_rows": 593, "request_coverage_fraction": 0.999321809965709, "requested_power_limit_w": 55.0, "start_temperature_c": 56.0, "terminal_rise_c": 12.0, "time_weighted_mean_power_w": 54.723754128096, "workload_phase": "decode" }, "phase_validation_pulse_01": { "completed_batch_count": 5, "completed_request_rows": 40, "duration_s": 59.948447462999866, "end_temperature_c": 69.0, "first_peak_time_s": 59.84963847100016, "first_request_start_offset_s": -0.026281500000095548, "integrated_energy_j": 3272.675364249615, "last_peak_time_s": 59.84963847100016, "last_request_end_offset_s": 3.1734457980001025, "maximum_realized_graphics_clock_mhz": 2040.0, "maximum_sampling_gap_s": 0.11724501000026066, "mean_utilization_percent": 99.41750841750842, "median_memory_clock_mhz": 6251.0, "median_realized_graphics_clock_mhz": 705.0, "minimum_realized_graphics_clock_mhz": 585.0, "peak_rise_c": 16.0, "peak_sample_count": 1, "peak_temperature_c": 70.0, "raw_rows": 594, "request_coverage_fraction": 0.9998456175699073, "requested_power_limit_w": 55.0, "start_temperature_c": 54.0, "terminal_rise_c": 15.0, "time_weighted_mean_power_w": 54.59149490517678, "workload_phase": "prefill" } } }, "method": "pre-registered one-second level fits using measured power; exact-discrete free-run trajectories; no temperature finite differences", "model_comparison": { "phase_gain_minus_power_only_training_rmse_c": -0.5699168839735372, "phase_gain_minus_power_only_validation_rmse_c": -0.5576435879437723, "validation_was_not_used_to_choose_or_refit_a_model": true }, "models": { "phase_gain_one_state_rc": { "acceptance": { "accepted_for_mixed_serving_thermal_constraints": false, "all_criteria_passed": false, "criteria": { "asymptotically_stable": { "passed": true, "required": true, "value": true }, "full_jacobian_rank": { "jacobian_numerical_rank": 4, "parameter_count": 4, "passed": true, "required": true }, "locally_identifiable": { "condition_number_limit": 100000000.0, "passed": true, "required": true, "value": true }, "optimizer_success": { "passed": true, "required": true, "value": true }, "physical_parameters": { "passed": true, "required": true, "value": true }, "replicated_multistart_solution": { "minimum_required_for_each": 2, "near_optimal_multistarts_within_one_percent": 10, "passed": true, "required": true, "successful_multistarts": 10 }, "validation_maximum_absolute_or_peak_error": { "maximum_absolute_per_pulse_peak_error_c": 1.1894546061778755, "maximum_absolute_trajectory_error_c": 2.7310039794374816, "operator": "<", "passed": false, "threshold_c": 1.0, "value_c": 2.7310039794374816 }, "validation_phase_contrast_error": { "absolute_error_c": 1.431289590989124, "contrast_definition": "prefill minus decode peak temperature rise on the separate one-second validation grids", "observed_contrast_c": 2.0, "operator": "<", "passed": false, "predicted_contrast_c": 3.431289590989124, "signed_error_c": 1.431289590989124, "threshold_c": 1.0 }, "validation_rmse": { "operator": "<", "passed": true, "threshold_c": 1.0, "value_c": 0.8696553321716621 } }, "rule": "all validation errors are strictly below 1 C and the fitted parameters are physical, stable, locally identified, full-rank, and reproduced by at least two successful near-optimal starts", "verdict": "rejected" }, "final_training_fit": { "diagnostics": { "asymptotically_stable": true, "continuous_poles_per_s": [ -0.0318869625164589 ], "discrete_poles_at_one_second": [ 0.9686160658137951 ], "jacobian_condition_number": 383.71782775965926, "jacobian_numerical_rank": 4, "jacobian_singular_values": [ 319.45296852419114, 112.22875911445492, 24.48138974043549, 0.8325205278819616 ], "locally_identifiable": true, "measured_power_span_w": 28.239967611099537, "minimum_transformed_bound_margin": 0.9613164586528914, "near_optimal_multistarts_within_one_percent": 10, "near_optimal_transformed_parameter_spread": [ 2.938227671478444e-05, 1.7511905670009043e-06, 2.5330167012782567e-06, 1.5659321020433659e-07 ], "optimizer_evaluations": 15, "optimizer_message": "`ftol` termination condition is satisfied.", "optimizer_success": true, "parameter_count": 4, "positive_effective_power_gain": true, "positive_thermal_parameters": true, "prefill_power_gain_bounds": [ 0.25, 3.0 ], "successful_multistarts": 10, "training_phase_counts": { "decode": 2, "prefill": 2 } }, "objective_sum_squared_c": 199.08673431958547, "parameters": { "ambient_temperature_c": 39.565264405145506, "beta": 0.14716746489653754, "prefill_power_gain": 1.1471674648965375, "thermal_capacitance_j_per_c": 60.7439434241958, "thermal_resistance_c_per_w": 0.5162782940900783, "thermal_time_constant_s": 31.360779487348037 } }, "training_evaluation": { "aggregate": { "absolute_peak_temperature_error_c": 0.3300028569910154, "mae_c": 0.6777106268982217, "maximum_absolute_error_c": 3.57323569770346, "maximum_absolute_per_pulse_peak_error_c": 2.327178649671488, "mean_absolute_per_pulse_peak_error_c": 0.8148697090716244, "mean_residual_c": -0.1528676043324373, "peak_temperature_error_c": 0.3300028569910154, "pulse_count": 4, "residual_autocorrelation": { "lag_10_s": 0.3851586565082845, "lag_1_s": 0.7594482568295176, "lag_30_s": -0.4326594954692055, "lag_5_s": 0.64678780121846 }, "rmse_c": 0.9184703899889092, "sample_count": 236 }, "per_pulse": { "phase_training_pulse_00": { "metrics": { "absolute_peak_temperature_error_c": 2.327178649671488, "mae_c": 0.6929539993478689, "maximum_absolute_error_c": 2.345832273171773, "mean_residual_c": 0.04799837746819383, "peak_temperature_error_c": -2.327178649671488, "residual_autocorrelation": { "lag_10_s": 0.3749162732875332, "lag_1_s": 0.7056586133971943, "lag_30_s": -0.39224085262559893, "lag_5_s": 0.5462359771759685 }, "rmse_c": 0.8271660765912097, "sample_count": 74 }, "requested_power_limit_w": 46.0, "scheduled_duration_s": 75.0, "workload_phase": "decode" }, "phase_training_pulse_01": { "metrics": { "absolute_peak_temperature_error_c": 0.3300028569910154, "mae_c": 1.049238070642173, "maximum_absolute_error_c": 3.57323569770346, "mean_residual_c": -0.63912230818908, "peak_temperature_error_c": 0.3300028569910154, "residual_autocorrelation": { "lag_10_s": 0.8120475744826366, "lag_1_s": 0.9265849932638685, "lag_30_s": -0.37949759276269873, "lag_5_s": 0.899126946591525 }, "rmse_c": 1.4616833245958438, "sample_count": 44 }, "requested_power_limit_w": 61.0, "scheduled_duration_s": 45.0, "workload_phase": "prefill" }, "phase_training_pulse_02": { "metrics": { "absolute_peak_temperature_error_c": 0.3303383582480848, "mae_c": 0.4762794154895177, "maximum_absolute_error_c": 1.75643032393031, "mean_residual_c": -0.06953399437356861, "peak_temperature_error_c": -0.3303383582480848, "residual_autocorrelation": { "lag_10_s": 0.3130616739386235, "lag_1_s": 0.7909492103160034, "lag_30_s": -0.46860084369339644, "lag_5_s": 0.6176094187232081 }, "rmse_c": 0.6047083099792162, "sample_count": 74 }, "requested_power_limit_w": 46.0, "scheduled_duration_s": 75.0, "workload_phase": "prefill" }, "phase_training_pulse_03": { "metrics": { "absolute_peak_temperature_error_c": 0.2719589713759092, "mae_c": 0.6193172759490477, "maximum_absolute_error_c": 1.9376835659166645, "mean_residual_c": -0.144584941162226, "peak_temperature_error_c": -0.2719589713759092, "residual_autocorrelation": { "lag_10_s": -0.12167334327159088, "lag_1_s": 0.23784798901716292, "lag_30_s": -0.3914826593779827, "lag_5_s": 0.2356125951206043 }, "rmse_c": 0.7889740544003814, "sample_count": 44 }, "requested_power_limit_w": 61.0, "scheduled_duration_s": 45.0, "workload_phase": "decode" } } }, "validation_evaluation": { "aggregate": { "absolute_peak_temperature_error_c": 0.24183498481124843, "mae_c": 0.6356571688854492, "maximum_absolute_error_c": 2.7310039794374816, "maximum_absolute_per_pulse_peak_error_c": 1.1894546061778755, "mean_absolute_per_pulse_peak_error_c": 0.715644795494562, "mean_residual_c": -0.23968855826985025, "peak_temperature_error_c": 0.24183498481124843, "pulse_count": 2, "residual_autocorrelation": { "lag_10_s": 0.44935952489567843, "lag_1_s": 0.6334436806775302, "lag_30_s": -0.42030937068661134, "lag_5_s": 0.7460526617993001 }, "rmse_c": 0.8696553321716621, "sample_count": 118 }, "evaluation_passes": 1, "matched_55w_decode_prefill": { "decode": { "block_id": "phase_validation_pulse_00", "metrics": { "absolute_peak_temperature_error_c": 1.1894546061778755, "mae_c": 0.520557114109162, "maximum_absolute_error_c": 2.114305337687881, "mean_residual_c": -0.26757864468056614, "peak_temperature_error_c": -1.1894546061778755, "residual_autocorrelation": { "lag_10_s": 0.14917326346868182, "lag_1_s": 0.17954036890943426, "lag_30_s": -0.4380451894208547, "lag_5_s": 0.5854017732608131 }, "rmse_c": 0.6869514834100526, "sample_count": 59 }, "observed_peak_rise_c": 13.0, "observed_peak_temperature_c": 67.0, "predicted_peak_rise_c": 11.810545393822125, "predicted_peak_temperature_c": 65.81054539382212 }, "matching_rule": "55 W requested cap, 60 s duration, cold start", "observed_prefill_minus_decode_peak_c": 2.0, "observed_prefill_minus_decode_peak_rise_c": 2.0, "predicted_prefill_minus_decode_peak_c": 3.431289590989124, "predicted_prefill_minus_decode_peak_rise_c": 3.431289590989124, "prefill": { "block_id": "phase_validation_pulse_01", "metrics": { "absolute_peak_temperature_error_c": 0.24183498481124843, "mae_c": 0.7507572236617366, "maximum_absolute_error_c": 2.7310039794374816, "mean_residual_c": -0.21179847185913436, "peak_temperature_error_c": 0.24183498481124843, "residual_autocorrelation": { "lag_10_s": 0.6801954658119723, "lag_1_s": 0.825644728155473, "lag_30_s": -0.23021727891780228, "lag_5_s": 0.8344780018210952 }, "rmse_c": 1.0201462899946931, "sample_count": 59 }, "observed_peak_rise_c": 15.0, "observed_peak_temperature_c": 69.0, "predicted_peak_rise_c": 15.241834984811248, "predicted_peak_temperature_c": 69.24183498481125 } }, "per_pulse": { "phase_validation_pulse_00": { "metrics": { "absolute_peak_temperature_error_c": 1.1894546061778755, "mae_c": 0.520557114109162, "maximum_absolute_error_c": 2.114305337687881, "mean_residual_c": -0.26757864468056614, "peak_temperature_error_c": -1.1894546061778755, "residual_autocorrelation": { "lag_10_s": 0.14917326346868182, "lag_1_s": 0.17954036890943426, "lag_30_s": -0.4380451894208547, "lag_5_s": 0.5854017732608131 }, "rmse_c": 0.6869514834100526, "sample_count": 59 }, "requested_power_limit_w": 55.0, "scheduled_duration_s": 60.0, "workload_phase": "decode" }, "phase_validation_pulse_01": { "metrics": { "absolute_peak_temperature_error_c": 0.24183498481124843, "mae_c": 0.7507572236617366, "maximum_absolute_error_c": 2.7310039794374816, "mean_residual_c": -0.21179847185913436, "peak_temperature_error_c": 0.24183498481124843, "residual_autocorrelation": { "lag_10_s": 0.6801954658119723, "lag_1_s": 0.825644728155473, "lag_30_s": -0.23021727891780228, "lag_5_s": 0.8344780018210952 }, "rmse_c": 1.0201462899946931, "sample_count": 59 }, "requested_power_limit_w": 55.0, "scheduled_duration_s": 60.0, "workload_phase": "prefill" } } } }, "power_only_one_state_rc": { "acceptance": { "accepted_for_mixed_serving_thermal_constraints": false, "all_criteria_passed": false, "criteria": { "asymptotically_stable": { "passed": true, "required": true, "value": true }, "full_jacobian_rank": { "jacobian_numerical_rank": 3, "parameter_count": 3, "passed": true, "required": true }, "locally_identifiable": { "condition_number_limit": 100000000.0, "passed": true, "required": true, "value": true }, "optimizer_success": { "passed": true, "required": true, "value": true }, "physical_parameters": { "passed": true, "required": true, "value": true }, "replicated_multistart_solution": { "minimum_required_for_each": 2, "near_optimal_multistarts_within_one_percent": 10, "passed": true, "required": true, "successful_multistarts": 10 }, "validation_maximum_absolute_or_peak_error": { "maximum_absolute_per_pulse_peak_error_c": 1.1228406146562264, "maximum_absolute_trajectory_error_c": 2.8851109832398976, "operator": "<", "passed": false, "threshold_c": 1.0, "value_c": 2.8851109832398976 }, "validation_phase_contrast_error": { "absolute_error_c": 2.103189848480042, "contrast_definition": "prefill minus decode peak temperature rise on the separate one-second validation grids", "observed_contrast_c": 2.0, "operator": "<", "passed": false, "predicted_contrast_c": -0.1031898484800422, "signed_error_c": -2.103189848480042, "threshold_c": 1.0 }, "validation_rmse": { "operator": "<", "passed": false, "threshold_c": 1.0, "value_c": 1.4272989201154345 } }, "rule": "all validation errors are strictly below 1 C and the fitted parameters are physical, stable, locally identified, full-rank, and reproduced by at least two successful near-optimal starts", "verdict": "rejected" }, "final_training_fit": { "diagnostics": { "asymptotically_stable": true, "continuous_poles_per_s": [ -0.027371588697128024 ], "discrete_poles_at_one_second": [ 0.9729996186804056 ], "jacobian_condition_number": 402.6688924636258, "jacobian_numerical_rank": 3, "jacobian_singular_values": [ 283.6011917611884, 23.946932552621675, 0.7043037022950732 ], "locally_identifiable": true, "measured_power_span_w": 28.239967611099537, "minimum_transformed_bound_margin": 2.7791486090646016, "near_optimal_multistarts_within_one_percent": 10, "near_optimal_transformed_parameter_spread": [ 7.673399359475752e-05, 3.555262910015511e-06, 4.250347563239387e-06 ], "optimizer_evaluations": 19, "optimizer_message": "`ftol` termination condition is satisfied.", "optimizer_success": true, "parameter_count": 3, "positive_thermal_parameters": true, "successful_multistarts": 10 }, "objective_sum_squared_c": 522.8100158412334, "parameters": { "ambient_temperature_c": 37.34514311778845, "thermal_capacitance_j_per_c": 58.839490021295305, "thermal_resistance_c_per_w": 0.6209134889138616, "thermal_time_constant_s": 36.53423303503481 } }, "training_evaluation": { "aggregate": { "absolute_peak_temperature_error_c": 1.180777033656753, "mae_c": 1.3096144338046691, "maximum_absolute_error_c": 3.656166289255978, "maximum_absolute_per_pulse_peak_error_c": 1.6994142857170118, "mean_absolute_per_pulse_peak_error_c": 1.226040289093472, "mean_residual_c": -0.2023489851680041, "peak_temperature_error_c": -1.180777033656753, "pulse_count": 4, "residual_autocorrelation": { "lag_10_s": 0.8282221981983825, "lag_1_s": 0.9117619226894861, "lag_30_s": 0.5589980012360515, "lag_5_s": 0.8849936892817001 }, "rmse_c": 1.4883872739624464, "sample_count": 236 }, "per_pulse": { "phase_training_pulse_00": { "metrics": { "absolute_peak_temperature_error_c": 0.42764190858825657, "mae_c": 1.1980967775237823, "maximum_absolute_error_c": 2.3035293505014423, "mean_residual_c": 1.086009365632016, "peak_temperature_error_c": -0.42764190858825657, "residual_autocorrelation": { "lag_10_s": 0.28678750209871096, "lag_1_s": 0.6813785260919679, "lag_30_s": -0.44832880955526017, "lag_5_s": 0.49986683039607566 }, "rmse_c": 1.3487155351661482, "sample_count": 74 }, "requested_power_limit_w": 46.0, "scheduled_duration_s": 75.0, "workload_phase": "decode" }, "phase_training_pulse_01": { "metrics": { "absolute_peak_temperature_error_c": 1.180777033656753, "mae_c": 1.8010806228301723, "maximum_absolute_error_c": 3.656166289255978, "mean_residual_c": -1.8010806228301723, "peak_temperature_error_c": -1.180777033656753, "residual_autocorrelation": { "lag_10_s": 0.6844914940053667, "lag_1_s": 0.8516456835734085, "lag_30_s": -0.5222860814243605, "lag_5_s": 0.8133141279117726 }, "rmse_c": 2.021755294975409, "sample_count": 44 }, "requested_power_limit_w": 61.0, "scheduled_duration_s": 45.0, "workload_phase": "prefill" }, "phase_training_pulse_02": { "metrics": { "absolute_peak_temperature_error_c": 1.596327928411867, "mae_c": 1.169357410699736, "maximum_absolute_error_c": 2.0932794809161663, "mean_residual_c": -1.169357410699736, "peak_temperature_error_c": -1.596327928411867, "residual_autocorrelation": { "lag_10_s": 0.25351031303147364, "lag_1_s": 0.6267444670353877, "lag_30_s": -0.2929529878727606, "lag_5_s": 0.4485096330816267 }, "rmse_c": 1.2494454511224107, "sample_count": 74 }, "requested_power_limit_w": 46.0, "scheduled_duration_s": 75.0, "workload_phase": "prefill" }, "phase_training_pulse_03": { "metrics": { "absolute_peak_temperature_error_c": 1.6994142857170118, "mae_c": 1.241587478291682, "maximum_absolute_error_c": 2.45915189802092, "mean_residual_c": 0.8559305054520432, "peak_temperature_error_c": 1.6994142857170118, "residual_autocorrelation": { "lag_10_s": 0.4656651895085471, "lag_1_s": 0.6636161939394328, "lag_30_s": -0.23146731135143495, "lag_5_s": 0.6286183507441672 }, "rmse_c": 1.452500522689621, "sample_count": 44 }, "requested_power_limit_w": 61.0, "scheduled_duration_s": 45.0, "workload_phase": "decode" } } }, "validation_evaluation": { "aggregate": { "absolute_peak_temperature_error_c": 1.0196507661761842, "mae_c": 1.2816417144849135, "maximum_absolute_error_c": 2.8851109832398976, "maximum_absolute_per_pulse_peak_error_c": 1.1228406146562264, "mean_absolute_per_pulse_peak_error_c": 1.051594924240021, "mean_residual_c": -0.24567039879188549, "peak_temperature_error_c": -1.0196507661761842, "pulse_count": 2, "residual_autocorrelation": { "lag_10_s": 0.8513328487489752, "lag_1_s": 0.8725839470045729, "lag_30_s": 0.7119608658323165, "lag_5_s": 0.919598936454576 }, "rmse_c": 1.4272989201154345, "sample_count": 118 }, "evaluation_passes": 1, "matched_55w_decode_prefill": { "decode": { "block_id": "phase_validation_pulse_00", "metrics": { "absolute_peak_temperature_error_c": 0.9803492338238158, "mae_c": 1.1928360303412524, "maximum_absolute_error_c": 2.2446786843485995, "mean_residual_c": 0.8791066010448036, "peak_temperature_error_c": 0.9803492338238158, "residual_autocorrelation": { "lag_10_s": 0.5966036338616748, "lag_1_s": 0.6555331748377423, "lag_30_s": -0.05054771976139612, "lag_5_s": 0.8185730176499005 }, "rmse_c": 1.3245540317756705, "sample_count": 59 }, "observed_peak_rise_c": 13.0, "observed_peak_temperature_c": 67.0, "predicted_peak_rise_c": 13.980349233823816, "predicted_peak_temperature_c": 67.98034923382382 }, "matching_rule": "55 W requested cap, 60 s duration, cold start", "observed_prefill_minus_decode_peak_c": 2.0, "observed_prefill_minus_decode_peak_rise_c": 2.0, "predicted_prefill_minus_decode_peak_c": -0.1031898484800422, "predicted_prefill_minus_decode_peak_rise_c": -0.1031898484800422, "prefill": { "block_id": "phase_validation_pulse_01", "metrics": { "absolute_peak_temperature_error_c": 1.1228406146562264, "mae_c": 1.3704473986285746, "maximum_absolute_error_c": 2.8851109832398976, "mean_residual_c": -1.3704473986285746, "peak_temperature_error_c": -1.1228406146562264, "residual_autocorrelation": { "lag_10_s": 0.46522441347782295, "lag_1_s": 0.6142897059579316, "lag_30_s": -0.21615376803282396, "lag_5_s": 0.6770376221182488 }, "rmse_c": 1.523128698315473, "sample_count": 59 }, "observed_peak_rise_c": 15.0, "observed_peak_temperature_c": 69.0, "predicted_peak_rise_c": 13.877159385343774, "predicted_peak_temperature_c": 67.87715938534377 } }, "per_pulse": { "phase_validation_pulse_00": { "metrics": { "absolute_peak_temperature_error_c": 0.9803492338238158, "mae_c": 1.1928360303412524, "maximum_absolute_error_c": 2.2446786843485995, "mean_residual_c": 0.8791066010448036, "peak_temperature_error_c": 0.9803492338238158, "residual_autocorrelation": { "lag_10_s": 0.5966036338616748, "lag_1_s": 0.6555331748377423, "lag_30_s": -0.05054771976139612, "lag_5_s": 0.8185730176499005 }, "rmse_c": 1.3245540317756705, "sample_count": 59 }, "requested_power_limit_w": 55.0, "scheduled_duration_s": 60.0, "workload_phase": "decode" }, "phase_validation_pulse_01": { "metrics": { "absolute_peak_temperature_error_c": 1.1228406146562264, "mae_c": 1.3704473986285746, "maximum_absolute_error_c": 2.8851109832398976, "mean_residual_c": -1.3704473986285746, "peak_temperature_error_c": -1.1228406146562264, "residual_autocorrelation": { "lag_10_s": 0.46522441347782295, "lag_1_s": 0.6142897059579316, "lag_30_s": -0.21615376803282396, "lag_5_s": 0.6770376221182488 }, "rmse_c": 1.523128698315473, "sample_count": 59 }, "requested_power_limit_w": 55.0, "scheduled_duration_s": 60.0, "workload_phase": "prefill" } } } } }, "observed_validation_contrast": { "one_second_aligned": { "alignment_rule": "separate grids anchored at each pulse's first telemetry row", "decode": { "duration_s": 59.0, "end_temperature_c": 66.0, "peak_rise_c": 13.0, "peak_temperature_c": 67.0, "start_temperature_c": 54.0, "temperature_states": 60, "terminal_rise_c": 12.0 }, "prefill": { "duration_s": 59.0, "end_temperature_c": 69.0, "peak_rise_c": 15.0, "peak_temperature_c": 69.0, "start_temperature_c": 54.0, "temperature_states": 60, "terminal_rise_c": 15.0 }, "prefill_minus_decode_peak_c": 2.0, "prefill_minus_decode_peak_rise_c": 2.0, "prefill_minus_decode_terminal_rise_c": 3.0 }, "raw_telemetry": { "decode": { "duration_s": 59.92533942399996, "end_temperature_c": 68.0, "first_peak_time_s": 57.09476940700006, "integrated_energy_j": 3279.3395406816717, "last_peak_time_s": 59.92533942399996, "maximum_realized_graphics_clock_mhz": 1320.0, "maximum_sampling_gap_s": 0.1218580810000276, "mean_utilization_percent": 98.90050590219224, "median_memory_clock_mhz": 6251.0, "median_realized_graphics_clock_mhz": 345.0, "minimum_realized_graphics_clock_mhz": 270.0, "peak_rise_c": 12.0, "peak_sample_count": 5, "peak_temperature_c": 68.0, "raw_rows": 593, "start_temperature_c": 56.0, "terminal_rise_c": 12.0, "time_weighted_mean_power_w": 54.723754128096 }, "prefill": { "duration_s": 59.948447462999866, "end_temperature_c": 69.0, "first_peak_time_s": 59.84963847100016, "integrated_energy_j": 3272.675364249615, "last_peak_time_s": 59.84963847100016, "maximum_realized_graphics_clock_mhz": 2040.0, "maximum_sampling_gap_s": 0.11724501000026066, "mean_utilization_percent": 99.41750841750842, "median_memory_clock_mhz": 6251.0, "median_realized_graphics_clock_mhz": 705.0, "minimum_realized_graphics_clock_mhz": 585.0, "peak_rise_c": 16.0, "peak_sample_count": 1, "peak_temperature_c": 70.0, "raw_rows": 594, "start_temperature_c": 54.0, "terminal_rise_c": 15.0, "time_weighted_mean_power_w": 54.59149490517678 }, "prefill_minus_decode_peak_c": 2.0, "prefill_minus_decode_peak_rise_c": 4.0, "prefill_minus_decode_terminal_rise_c": 3.0 } }, "pre_registration": { "candidate_models": [ "power_only_one_state_rc", "phase_gain_one_state_rc" ], "maximum_optimizer_evaluations": 2500, "multistart_count": 10, "phase_gain_equation": "P_eff = P * (1 + beta * I_prefill)", "prefill_power_gain_bounds": [ 0.25, 3.0 ], "random_seed": 20260902, "training_pulse_count": 4, "training_pulse_ids": [ "phase_training_pulse_00", "phase_training_pulse_01", "phase_training_pulse_02", "phase_training_pulse_03" ], "training_sequence": "phase_training", "untouched_validation_sequence": "phase_validation", "validation_pulse_count": 2, "validation_pulse_ids": [ "phase_validation_pulse_00", "phase_validation_pulse_01" ], "validation_used_for_fit_or_model_selection": false }, "preprocessing": { "bin_width_s": 1.0, "finite_differencing": false, "power_statistic": "time-weighted mean of measured board power", "pulse_initialization": "verified post-relock junction temperature recorded immediately before each pulse", "pulse_samples": { "phase_training_pulse_00": { "binned_duration_s": 74.0, "binned_temperature_states": 75, "measured_power_intervals": 74, "raw_rows": 749, "requested_power_limit_w": 46.0, "split": "training", "workload_phase": "decode" }, "phase_training_pulse_01": { "binned_duration_s": 43.999999999999886, "binned_temperature_states": 45, "measured_power_intervals": 44, "raw_rows": 449, "requested_power_limit_w": 61.0, "split": "training", "workload_phase": "prefill" }, "phase_training_pulse_02": { "binned_duration_s": 74.0, "binned_temperature_states": 75, "measured_power_intervals": 74, "raw_rows": 747, "requested_power_limit_w": 46.0, "split": "training", "workload_phase": "prefill" }, "phase_training_pulse_03": { "binned_duration_s": 44.0, "binned_temperature_states": 45, "measured_power_intervals": 44, "raw_rows": 448, "requested_power_limit_w": 61.0, "split": "training", "workload_phase": "decode" }, "phase_validation_pulse_00": { "binned_duration_s": 59.0, "binned_temperature_states": 60, "measured_power_intervals": 59, "raw_rows": 593, "requested_power_limit_w": 55.0, "split": "validation", "workload_phase": "decode" }, "phase_validation_pulse_01": { "binned_duration_s": 59.0, "binned_temperature_states": 60, "measured_power_intervals": 59, "raw_rows": 594, "requested_power_limit_w": 55.0, "split": "validation", "workload_phase": "prefill" } }, "temperature_statistic": "nearest integer sensor reading at each bin boundary" }, "schema_version": 1, "source": { "manifest": "thermal_phase_manifest.json", "request_row_count": 640, "requests": "l4_thermal_phase_requests.csv", "requests_sha256": "8f0c8628bebff01c6264428b7ae6ae668ca8bdcbbd3e010048f83e1179460be0", "source_git_revision": "0fdb4284ee509c4696968630c2b7b54eac3dd52e", "source_manifest_schema_version": 2, "telemetry": "l4_thermal_phase_telemetry.csv", "telemetry_row_count": 29171, "telemetry_sha256": "cb896ac8d06dc1ed8b0e4abea826a802b3cb9df9a35bce602c61284030be09f5" } }