[ { "50.00 percentile latency (ns)": 40540772262, "90.00 percentile latency (ns)": 72636904696, "95.00 percentile latency (ns)": 76646310127, "97.00 percentile latency (ns)": 78246030469, "99.00 percentile latency (ns)": 79855474850, "99.90 percentile latency (ns)": 80561174134, "Max latency (ns)": 80713667170, "Mean latency (ns)": 40551824369, "Min duration satisfied": "Yes", "Min latency (ns)": 96524089, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "sut", "Samples per second": 130833, "Scenario": "offline", "accelerator_memory_capacity": "16 GB per accelerator", "accelerator_memory_configuration": "HBM", "accelerator_model_name": "tpu-v3.8", "accelerators_per_node": 32, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "characteristics.accuracy": 76.468, "characteristics.good": 38234, "characteristics.samples_per_second": 130833, "characteristics.samples_per_second.normalized_per_core": 4088.53125, "characteristics.samples_per_second.normalized_per_processor": 4088.53125, "characteristics.total": 50000, "ck_system": "tpu-v3.8x4", "ck_used": true, "dataset": "ImageNet 2012", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/imagenet2012.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.accuracy", "dim_y_maximize": true, "division": "closed", "formal_model": "resnet50-v1.5", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "framework": "tf-nightly==1.15.0-dev20190730", "host_memory_capacity": "360 GB", "host_processor_core_count": 120, "host_processor_model_name": "AMD EPYC 7V13 64-Core Processor", "host_processors_per_node": 2, "host_storage_capacity": "200 GB + 50 GB", "host_storage_type": "SSD", "hw_notes": "'Number_of_nodes' is the number of user programmable nodes. Each GCP TPU v3-8 has a non-user programmable 'tensorflow host' node which provides direct network and storage access to the TPU.", "informal_model": "resnet", "input_data_types": "bfloat16", "key.accuracy": "characteristics.accuracy", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 60000, "min_query_count": 1, "mlperf_version": 0.5, "normalize_cores": 32, "normalize_processors": 32, "note_code": "https://github.com/mlcommons/inference_results_v0.5/tree/master/closed/Google/code", "note_details": "https://github.com/mlcommons/inference_results_v0.5/tree/master/closed/Google/results/tpu-v3.8x4", "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": "TPU version 1.15.dev20190821; GCC 7.5.0; Python 3.7.10", "performance_issue_same": true, "performance_issue_same_index": 0, "performance_issue_unique": true, "performance_sample_count": 3072, "print_timestamps": true, "problem": false, "qsl_rng_seed": 3133965575612453542, "retraining": "None", "sample_index_rng_seed": 665484352860916858, "samples_per_query": 10560000, "schedule_rng_seed": 3622009729038561421, "starting_weights_filename": "https://zenodo.org/record/2535873/files/resnet50_v1.pb", "status": "available", "submitter": "Google", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Google", "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/tpu-v3.8x4", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "target_latency (ns)": 0, "target_qps": 160000, "task": "image classification", "task2": "image classification", "total_cores": 240, "uid": "ba24fa184ed25a64", "use_accelerator": true, "weight_data_types": "bfloat16", "weight_transformations": "None" }, { "50.00 percentile latency (ns)": 41127516115, "90.00 percentile latency (ns)": 73246796743, "95.00 percentile latency (ns)": 77260071975, "97.00 percentile latency (ns)": 78868338599, "99.00 percentile latency (ns)": 80476829972, "99.90 percentile latency (ns)": 81202868568, "Max latency (ns)": 81347092763, "Mean latency (ns)": 41032852918, "Min duration satisfied": "Yes", "Min latency (ns)": 85792149, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "sut", "Samples per second": 1038510.0, "Scenario": "offline", "accelerator_memory_capacity": "16 GB per accelerator", "accelerator_memory_configuration": "HBM", "accelerator_model_name": "tpu-v3.8", "accelerators_per_node": 256, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "characteristics.accuracy": 76.468, "characteristics.good": 38234, "characteristics.samples_per_second": 1038510.0, "characteristics.samples_per_second.normalized_per_core": 4056.6796875, "characteristics.samples_per_second.normalized_per_processor": 4056.6796875, "characteristics.total": 50000, "ck_system": "tpu-v3.8x32", "ck_used": true, "dataset": "ImageNet 2012", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/imagenet2012.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.accuracy", "dim_y_maximize": true, "division": "closed", "formal_model": "resnet50-v1.5", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "framework": "tf-nightly==1.15.0-dev20190730", "host_memory_capacity": "360 GB", "host_processor_core_count": 120, "host_processor_model_name": "AMD EPYC 7V13 64-Core Processor", "host_processors_per_node": 2, "host_storage_capacity": "200 GB + 50 GB", "host_storage_type": "SSD", "hw_notes": "'Number_of_nodes' is the number of user programmable nodes. Each GCP TPU v3-8 has a non-user programmable 'tensorflow host' node which provides direct network and storage access to the TPU.", "informal_model": "resnet", "input_data_types": "bfloat16", "key.accuracy": "characteristics.accuracy", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 60000, "min_query_count": 1, "mlperf_version": 0.5, "normalize_cores": 256, "normalize_processors": 256, "note_code": "https://github.com/mlcommons/inference_results_v0.5/tree/master/closed/Google/code", "note_details": "https://github.com/mlcommons/inference_results_v0.5/tree/master/closed/Google/results/tpu-v3.8x32", "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": "TPU version 1.15.dev20190821; GCC 7.5.0; Python 3.7.10", "performance_issue_same": true, "performance_issue_same_index": 0, "performance_issue_unique": true, "performance_sample_count": 3072, "print_timestamps": true, "problem": false, "qsl_rng_seed": 3133965575612453542, "retraining": "None", "sample_index_rng_seed": 665484352860916858, "samples_per_query": 84480000, "schedule_rng_seed": 3622009729038561421, "starting_weights_filename": "https://zenodo.org/record/2535873/files/resnet50_v1.pb", "status": "available", "submitter": "Google", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Google", "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/tpu-v3.8x32", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "target_latency (ns)": 0, "target_qps": 1280000.0, "task": "image classification", "task2": "image classification", "total_cores": 240, "uid": "4090853642428705", "use_accelerator": true, "weight_data_types": "bfloat16", "weight_transformations": "None" }, { "50.00 percentile latency (ns)": 40528594854, "90.00 percentile latency (ns)": 72609185757, "95.00 percentile latency (ns)": 76611954709, "97.00 percentile latency (ns)": 78217223713, "99.00 percentile latency (ns)": 79817596212, "99.90 percentile latency (ns)": 80539829626, "Max latency (ns)": 80737835249, "Mean latency (ns)": 40531004086, "Min duration satisfied": "Yes", "Min latency (ns)": 89067554, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "sut", "Samples per second": 261587, "Scenario": "offline", "accelerator_memory_capacity": "16 GB per accelerator", "accelerator_memory_configuration": "HBM", "accelerator_model_name": "tpu-v3.8", "accelerators_per_node": 64, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "characteristics.accuracy": 76.468, "characteristics.good": 38234, "characteristics.samples_per_second": 261587, "characteristics.samples_per_second.normalized_per_core": 4087.296875, "characteristics.samples_per_second.normalized_per_processor": 4087.296875, "characteristics.total": 50000, "ck_system": "tpu-v3.8x8", "ck_used": true, "dataset": "ImageNet 2012", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/imagenet2012.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.accuracy", "dim_y_maximize": true, "division": "closed", "formal_model": "resnet50-v1.5", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "framework": "tf-nightly==1.15.0-dev20190730", "host_memory_capacity": "360 GB", "host_processor_core_count": 120, "host_processor_model_name": "AMD EPYC 7V13 64-Core Processor", "host_processors_per_node": 2, "host_storage_capacity": "200 GB + 50 GB", "host_storage_type": "SSD", "hw_notes": "'Number_of_nodes' is the number of user programmable nodes. Each GCP TPU v3-8 has a non-user programmable 'tensorflow host' node which provides direct network and storage access to the TPU.", "informal_model": "resnet", "input_data_types": "bfloat16", "key.accuracy": "characteristics.accuracy", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 60000, "min_query_count": 1, "mlperf_version": 0.5, "normalize_cores": 64, "normalize_processors": 64, "note_code": "https://github.com/mlcommons/inference_results_v0.5/tree/master/closed/Google/code", "note_details": "https://github.com/mlcommons/inference_results_v0.5/tree/master/closed/Google/results/tpu-v3.8x8", "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": "TPU version 1.15.dev20190821; GCC 7.5.0; Python 3.7.10", "performance_issue_same": true, "performance_issue_same_index": 0, "performance_issue_unique": true, "performance_sample_count": 3072, "print_timestamps": true, "problem": false, "qsl_rng_seed": 3133965575612453542, "retraining": "None", "sample_index_rng_seed": 665484352860916858, "samples_per_query": 21120000, "schedule_rng_seed": 3622009729038561421, "starting_weights_filename": "https://zenodo.org/record/2535873/files/resnet50_v1.pb", "status": "available", "submitter": "Google", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Google", "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/tpu-v3.8x8", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "target_latency (ns)": 0, "target_qps": 320000, "task": "image classification", "task2": "image classification", "total_cores": 240, "uid": "92de5294f949d972", "use_accelerator": true, "weight_data_types": "bfloat16", "weight_transformations": "None" }, { "50.00 percentile latency (ns)": 40587754739, "90.00 percentile latency (ns)": 72633376243, "95.00 percentile latency (ns)": 76640886389, "97.00 percentile latency (ns)": 78233787775, "99.00 percentile latency (ns)": 79839333549, "99.90 percentile latency (ns)": 80561119225, "Max latency (ns)": 80695165849, "Mean latency (ns)": 40576243360, "Min duration satisfied": "Yes", "Min latency (ns)": 403096095, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "sut", "Samples per second": 65431.4, "Scenario": "offline", "accelerator_memory_capacity": "16 GB per accelerator", "accelerator_memory_configuration": "HBM", "accelerator_model_name": "tpu-v3.8", "accelerators_per_node": 16, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "characteristics.accuracy": 76.468, "characteristics.good": 38234, "characteristics.samples_per_second": 65431.4, "characteristics.samples_per_second.normalized_per_core": 4089.4625, "characteristics.samples_per_second.normalized_per_processor": 4089.4625, "characteristics.total": 50000, "ck_system": "tpu-v3.8x2", "ck_used": true, "dataset": "ImageNet 2012", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/imagenet2012.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.accuracy", "dim_y_maximize": true, "division": "closed", "formal_model": "resnet50-v1.5", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "framework": "tf-nightly==1.15.0-dev20190730", "host_memory_capacity": "360 GB", "host_processor_core_count": 120, "host_processor_model_name": "AMD EPYC 7V13 64-Core Processor", "host_processors_per_node": 2, "host_storage_capacity": "200 GB + 50 GB", "host_storage_type": "SSD", "hw_notes": "'Number_of_nodes' is the number of user programmable nodes. Each GCP TPU v3-8 has a non-user programmable 'tensorflow host' node which provides direct network and storage access to the TPU.", "informal_model": "resnet", "input_data_types": "bfloat16", "key.accuracy": "characteristics.accuracy", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 60000, "min_query_count": 1, "mlperf_version": 0.5, "normalize_cores": 16, "normalize_processors": 16, "note_code": "https://github.com/mlcommons/inference_results_v0.5/tree/master/closed/Google/code", "note_details": "https://github.com/mlcommons/inference_results_v0.5/tree/master/closed/Google/results/tpu-v3.8x2", "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": "TPU version 1.15.dev20190821; GCC 7.5.0; Python 3.7.10", "performance_issue_same": true, "performance_issue_same_index": 0, "performance_issue_unique": true, "performance_sample_count": 3072, "print_timestamps": true, "problem": false, "qsl_rng_seed": 3133965575612453542, "retraining": "None", "sample_index_rng_seed": 665484352860916858, "samples_per_query": 5280000, "schedule_rng_seed": 3622009729038561421, "starting_weights_filename": "https://zenodo.org/record/2535873/files/resnet50_v1.pb", "status": "available", "submitter": "Google", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Google", "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/tpu-v3.8x2", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "target_latency (ns)": 0, "target_qps": 80000, "task": "image classification", "task2": "image classification", "total_cores": 240, "uid": "9bb6318b41aba8e4", "use_accelerator": true, "weight_data_types": "bfloat16", "weight_transformations": "None" }, { "50.00 percentile latency (ns)": 40308081257, "90.00 percentile latency (ns)": 72374164733, "95.00 percentile latency (ns)": 76383361466, "97.00 percentile latency (ns)": 77986434282, "99.00 percentile latency (ns)": 79592254453, "99.90 percentile latency (ns)": 80315082738, "Max latency (ns)": 80460510993, "Mean latency (ns)": 40311029963, "Min duration satisfied": "Yes", "Min latency (ns)": 86175860, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "sut", "Samples per second": 524978, "Scenario": "offline", "accelerator_memory_capacity": "16 GB per accelerator", "accelerator_memory_configuration": "HBM", "accelerator_model_name": "tpu-v3.8", "accelerators_per_node": 128, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "characteristics.accuracy": 76.468, "characteristics.good": 38234, "characteristics.samples_per_second": 524978, "characteristics.samples_per_second.normalized_per_core": 4101.390625, "characteristics.samples_per_second.normalized_per_processor": 4101.390625, "characteristics.total": 50000, "ck_system": "tpu-v3.8x16", "ck_used": true, "dataset": "ImageNet 2012", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/imagenet2012.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.accuracy", "dim_y_maximize": true, "division": "closed", "formal_model": "resnet50-v1.5", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "framework": "tf-nightly==1.15.0-dev20190730", "host_memory_capacity": "360 GB", "host_processor_core_count": 120, "host_processor_model_name": "AMD EPYC 7V13 64-Core Processor", "host_processors_per_node": 2, "host_storage_capacity": "200 GB + 50 GB", "host_storage_type": "SSD", "hw_notes": "'Number_of_nodes' is the number of user programmable nodes. Each GCP TPU v3-8 has a non-user programmable 'tensorflow host' node which provides direct network and storage access to the TPU.", "informal_model": "resnet", "input_data_types": "bfloat16", "key.accuracy": "characteristics.accuracy", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 60000, "min_query_count": 1, "mlperf_version": 0.5, "normalize_cores": 128, "normalize_processors": 128, "note_code": "https://github.com/mlcommons/inference_results_v0.5/tree/master/closed/Google/code", "note_details": "https://github.com/mlcommons/inference_results_v0.5/tree/master/closed/Google/results/tpu-v3.8x16", "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": "TPU version 1.15.dev20190821; GCC 7.5.0; Python 3.7.10", "performance_issue_same": true, "performance_issue_same_index": 0, "performance_issue_unique": true, "performance_sample_count": 3072, "print_timestamps": true, "problem": false, "qsl_rng_seed": 3133965575612453542, "retraining": "None", "sample_index_rng_seed": 665484352860916858, "samples_per_query": 42240000, "schedule_rng_seed": 3622009729038561421, "starting_weights_filename": "https://zenodo.org/record/2535873/files/resnet50_v1.pb", "status": "available", "submitter": "Google", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Google", "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/tpu-v3.8x16", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "target_latency (ns)": 0, "target_qps": 640000, "task": "image classification", "task2": "image classification", "total_cores": 240, "uid": "1d5013f39607441f", "use_accelerator": true, "weight_data_types": "bfloat16", "weight_transformations": "None" }, { "50.00 percentile latency (ns)": 40547295047, "90.00 percentile latency (ns)": 72608873966, "95.00 percentile latency (ns)": 76629641063, "97.00 percentile latency (ns)": 78246849845, "99.00 percentile latency (ns)": 79842477736, "99.90 percentile latency (ns)": 80564959996, "Max latency (ns)": 80694349670, "Mean latency (ns)": 40551893222, "Min duration satisfied": "Yes", "Min latency (ns)": 421051255, "Min queries satisfied": "Yes", "Mode": "Performance", "Result is": "VALID", "SUT name": "sut", "Samples per second": 32716, "Scenario": "offline", "accelerator_memory_capacity": "16 GB per accelerator", "accelerator_memory_configuration": "HBM", "accelerator_model_name": "tpu-v3.8", "accelerators_per_node": 8, "accuracy_log_probability": 0, "accuracy_log_rng_seed": 0, "characteristics.accuracy": 76.468, "characteristics.good": 38234, "characteristics.samples_per_second": 32716, "characteristics.samples_per_second.normalized_per_core": 4089.5, "characteristics.samples_per_second.normalized_per_processor": 4089.5, "characteristics.total": 50000, "ck_system": "tpu-v3.8x1", "ck_used": true, "dataset": "ImageNet 2012", "dataset_link": "https://github.com/ctuning/ck/blob/master/docs/mlperf-automation/datasets/imagenet2012.md", "dim_x_default": "characteristics.samples_per_second", "dim_x_maximize": true, "dim_y_default": "characteristics.accuracy", "dim_y_maximize": true, "division": "closed", "formal_model": "resnet50-v1.5", "formal_model_accuracy": 99.0, "formal_model_link": "https://github.com/mlcommons/ck-mlops/tree/main/package", "framework": "tf-nightly==1.15.0-dev20190730", "host_memory_capacity": "360 GB", "host_processor_core_count": 120, "host_processor_model_name": "AMD EPYC 7V13 64-Core Processor", "host_processors_per_node": 2, "host_storage_capacity": "200 GB + 50 GB", "host_storage_type": "SSD", "hw_notes": "'Number_of_nodes' is the number of user programmable nodes. Each GCP TPU v3-8 has a non-user programmable 'tensorflow host' node which provides direct network and storage access to the TPU.", "informal_model": "resnet", "input_data_types": "bfloat16", "key.accuracy": "characteristics.accuracy", "max_async_queries": 1, "max_duration (ms)": 0, "max_query_count": 0, "min_duration (ms)": 60000, "min_query_count": 1, "mlperf_version": 0.5, "normalize_cores": 8, "normalize_processors": 8, "note_code": "https://github.com/mlcommons/inference_results_v0.5/tree/master/closed/Google/code", "note_details": "https://github.com/mlcommons/inference_results_v0.5/tree/master/closed/Google/results/tpu-v3.8x1", "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": "TPU version 1.15.dev20190821; GCC 7.5.0; Python 3.7.10", "performance_issue_same": true, "performance_issue_same_index": 0, "performance_issue_unique": true, "performance_sample_count": 3072, "print_timestamps": true, "problem": false, "qsl_rng_seed": 3133965575612453542, "retraining": "None", "sample_index_rng_seed": 665484352860916858, "samples_per_query": 2640000, "schedule_rng_seed": 3622009729038561421, "starting_weights_filename": "https://zenodo.org/record/2535873/files/resnet50_v1.pb", "status": "available", "submitter": "Google", "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Google", "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/tpu-v3.8x1", "system_name": "Microsoft Corporation 7.0 (Virtual Machine)", "target_latency (ns)": 0, "target_qps": 40000, "task": "image classification", "task2": "image classification", "total_cores": 240, "uid": "f2db33ec4420b5df", "use_accelerator": true, "weight_data_types": "bfloat16", "weight_transformations": "None" } ]