[
  {
    "50.00 percentile latency (ns)": 49667145777,
    "90.00 percentile latency (ns)": 69504752443,
    "95.00 percentile latency (ns)": 71015167305,
    "97.00 percentile latency (ns)": 71515095990,
    "99.00 percentile latency (ns)": 71990360328,
    "99.90 percentile latency (ns)": 72128853511,
    "Max latency (ns)": 72147022561,
    "Mean latency (ns)": 45953543353,
    "Min duration satisfied": "Yes",
    "Min latency (ns)": 2000874963,
    "Min queries satisfied": "Yes",
    "Mode": "Performance",
    "Result is": "VALID",
    "SUT name": "RNNT SERVER",
    "Samples per second": 5488.79,
    "Scenario": "offline",
    "accelerator_frequency": "",
    "accelerator_host_interconnect": "",
    "accelerator_interconnect": "",
    "accelerator_interconnect_topology": "",
    "accelerator_memory_capacity": "16 GB",
    "accelerator_memory_configuration": "GDDR6",
    "accelerator_model_name": "NVIDIA T4",
    "accelerator_on-chip_memories": "",
    "accelerators_per_node": 4,
    "accuracy_log_probability": 0,
    "accuracy_log_rng_seed": 0,
    "accuracy_log_sampling_target": 0,
    "characteristics.accuracy": 92.5567656542425,
    "characteristics.samples_per_second": 5488.79,
    "characteristics.samples_per_second.normalized_per_core": 1372.1975,
    "characteristics.samples_per_second.normalized_per_processor": 1372.1975,
    "characteristics.word error rate": 7.4432343457575145,
    "ck_system": "T4x4_TRT",
    "ck_used": true,
    "cooling": "",
    "dataset": "LibriSpeech",
    "dataset_link": "",
    "dim_x_default": "characteristics.samples_per_second",
    "dim_x_maximize": true,
    "dim_y_default": "characteristics.accuracy",
    "dim_y_maximize": true,
    "division": "closed",
    "formal_model": "rnn-t",
    "formal_model_accuracy": 99.0,
    "formal_model_link": "",
    "framework": "TensorRT 7.2, CUDA 11.0 Update 1",
    "host_memory_capacity": "256 GB",
    "host_memory_configuration": "",
    "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": "1 TB",
    "host_storage_type": "SATA HDD",
    "hw_notes": "ECC off",
    "informal_model": "rnnt",
    "input_data_types": "fp16",
    "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": 0.7,
    "normalize_cores": 4,
    "normalize_processors": 4,
    "note_code": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/Altos/code",
    "note_details": "https://github.com/mlcommons/inference_results_v0.7/tree/master/closed/Altos/results/T4x4_TRT",
    "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": "TensorRT 7.2, CUDA 11.0 Update 1, cuDNN 8.0.2, DALI 0.25.0; GCC 7.5.0; Python 3.7.10",
    "performance_issue_same": true,
    "performance_issue_same_index": 0,
    "performance_issue_unique": true,
    "performance_sample_count": 2513,
    "print_timestamps": true,
    "problem": false,
    "qsl_rng_seed": 12786827339337101903,
    "retraining": "N",
    "sample_index_rng_seed": 12640797754436136668,
    "samples_per_query": 396000,
    "schedule_rng_seed": 3135815929913719677,
    "starting_weights_filename": "DistributedDataParallel_1576581068.9962234-epoch-100.pt",
    "status": "available",
    "submitter": "ALTOS",
    "submitter_link": "https://github.com/ctuning/ck-mlperf-inference/tree/main/bench.mlperf.submitter/Altos",
    "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/T4x4_TRT",
    "system_name": "Microsoft Corporation 7.0 (Virtual Machine)",
    "system_type": "datacenter",
    "target_latency (ns)": 0,
    "target_qps": 6000,
    "task": "speech recognition",
    "task2": "speech recognition",
    "total_cores": 240,
    "uid": "40b12cce8897497f",
    "use_accelerator": true,
    "weight_data_types": "int8,fp16",
    "weight_transformations": "quantization, affine fusion"
  }
]