[ { "50.00 percentile latency (ns)": 451081251834, "90.00 percentile latency (ns)": 654712263290, "95.00 percentile latency (ns)": 672911524560, "97.00 percentile latency (ns)": 678320724286, "99.00 percentile latency (ns)": 682605250940, "99.90 percentile latency (ns)": 683694645334, "Max latency (ns)": 683941610052, "Mean latency (ns)": 422590802388, "Min duration satisfied": "Yes", "Min latency (ns)": 10661746496, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "BERT Offline", "Samples per second": 289.498, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "N/A", "accelerator_memory_configuration": "", "accelerator_model_name": "N/A", "accelerator_on-chip_memories": "", "accelerators_per_node": 0, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.samples_per_second": 289.498, "characteristics.samples_per_second.normalized_per_core": 3.618725, "characteristics.samples_per_second.normalized_per_processor": 144.749, "ck_system": "1-node-2S-ICX-PyTorch-INT8", "ck_used": false, "cooling": "Air", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_x_maximize": true, "dim_y_default": "characteristics.samples_per_second", "dim_y_maximize": false, "division": "closed", "formal_model": "bert", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "PyTorch", "host_memory_capacity": "1024GB", "host_memory_configuration": "8 slots / 64GB each / 3200 MT/s per socket", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 40, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "Intel(R) Xeon(R) Platinum 8380 CPU @ 2.30GHz", "host_processors_per_node": 2, "host_storage_capacity": "", "host_storage_type": "", "hw_notes": "Intel Devlopment Platform (2U Server)", "informal_model": "bert-99", "input_data_types": "No change.", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.1, "normalize_cores": 80, "normalize_processors": 2, "note_code": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Intel/code", "note_details": "https://github.com/mlcommons/inference_results_v1.1/tree/master/closed/Intel/results/1-node-2S-ICX-PyTorch-INT8", "number_of_nodes": 1, "operating_system": "Ubuntu 20.04.1 LTS", "other_software_stack": "5.4.0-45-generic", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 10833, "print_timestamps": 0, "problem": false, "qsl_rng_seed": 1624344308455410291, "retraining": "We don’t retrain the model weight.", "sample_index_rng_seed": 517984244576520566, "samples_per_query": 198000, "schedule_rng_seed": 10051496985653635065, "starting_weights_filename": "The original weight filename: https://zenodo.org/record/4792496/files/pytorch_model.bin", "status": "available", "submitter": "Intel", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Intel", "sw_notes": "", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/1-node-2S-ICX-PyTorch-INT8", "system_name": "1-node-2S-ICX-PyTorch-INT8", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 300, "task": "NLP", "task2": "nlp", "total_cores": 80, "uid": "a7743298f9f77d37", "use_accelerator": false, "weight_data_types": "int8", "weight_transformations": "We load int8 model." }, { "50.00 percentile latency (ns)": 603758219333, "90.00 percentile latency (ns)": 860080158387, "95.00 percentile latency (ns)": 882588120879, "97.00 percentile latency (ns)": 888191260979, "99.00 percentile latency (ns)": 892970462380, "99.90 percentile latency (ns)": 894243779971, "Max latency (ns)": 894244386197, "Mean latency (ns)": 565887440058, "Min duration satisfied": "Yes", "Min latency (ns)": 55912451299, "Min queries satisfied": "Yes", "Mode": "PerformanceOnly", "Result is": "VALID", "SUT name": "PySUT", "Samples per second": 221.416, "Scenario": "offline", "accelerator_frequency": "", "accelerator_host_interconnect": "", "accelerator_interconnect": "", "accelerator_interconnect_topology": "", "accelerator_memory_capacity": "N/A", "accelerator_memory_configuration": "", "accelerator_model_name": "N/A", "accelerator_on-chip_memories": "", "accelerators_per_node": 0, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "accuracy_log_sampling_target": 0, "characteristics.samples_per_second": 221.416, "characteristics.samples_per_second.normalized_per_core": 0.9225666666666666, "characteristics.samples_per_second.normalized_per_processor": 110.708, "ck_system": "1-node-2S-ICX-MXNet-INT8", "ck_used": true, "cooling": "", "dataset": "SQuAD v1.1", "dataset_link": "", "dim_x_default": "seq_number", "dim_x_maximize": true, "dim_y_default": "characteristics.samples_per_second", "dim_y_maximize": false, "division": "closed", "formal_model": "bert", "formal_model_accuracy": 99.0, "formal_model_link": "", "framework": "MXNet", "host_memory_capacity": "1024GB", "host_memory_configuration": "8 slots / 64GB each / 3200 MT/s per socket", "host_networking": "", "host_networking_topology": "", "host_processor_caches": "", "host_processor_core_count": 120, "host_processor_frequency": "", "host_processor_interconnect": "", "host_processor_model_name": "AMD EPYC 7V13 64-Core Processor", "host_processors_per_node": 2, "host_storage_capacity": "", "host_storage_type": "", "hw_notes": "", "informal_model": "bert-99", "input_data_types": "Int8.", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 600000, "min_query_count": 1, "mlperf_version": 1.0, "normalize_cores": 240, "normalize_processors": 2, "note_code": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/Intel/code", "note_details": "https://github.com/mlcommons/inference_results_v1.0/tree/master/closed/Intel/results/1-node-2S-ICX-MXNet-INT8", "number_of_nodes": 1, "operating_system": "Ubuntu 18.04.5 LTS (Linux-5.4.0-1055-azure-x86_64-with-Ubuntu-18.04-bionic)", "other_software_stack": "5.4.0-66-generic; GCC 7.5.0; Python 3.7.10", "performance_issue_same": 0, "performance_issue_same_index": 0, "performance_issue_unique": 0, "performance_sample_count": 10833, "print_timestamps": 0, "problem": false, "qsl_rng_seed": 7322528924094909334, "retraining": "We don’t retrain the model weight.", "sample_index_rng_seed": 1570999273408051088, "samples_per_query": 198000, "schedule_rng_seed": 3507442325620259414, "starting_weights_filename": "The original weight filename: https://zenodo.org/record/2592612/files/resnet50_v1.onnx", "status": "preview", "submitter": "Intel", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Intel", "sw_notes": "Powered by CK v2.5.8 (https://github.com/ctuning/ck)", "system_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.system/1-node-2S-ICX-MXNet-INT8", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "system_type": "datacenter", "target_latency (ns)": 0, "target_qps": 300, "task": "NLP", "task2": "nlp", "total_cores": 240, "uid": "411d31d534a74b40", "use_accelerator": false, "weight_data_types": "FP32 + Int8", "weight_transformations": "We transfer the weight from fp32 datatype in onnx file to mxnet files then quatize to fp32+int8 datatype file by Intel® Low Precision Optimization Tool." } ]