def main(_):
"""
Starting point of the application
"""
flags = PARSER.parse_args()
params = _cmd_params(flags)
tf.logging.set_verbosity(tf.logging.ERROR)
# Optimization flags
os.environ['CUDA_CACHE_DISABLE'] = '0'
os.environ['HOROVOD_GPU_ALLREDUCE'] = 'NCCL'
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'
os.environ['TF_GPU_THREAD_MODE'] = 'gpu_private'
os.environ['TF_USE_CUDNN_BATCHNORM_SPATIAL_PERSISTENT'] = '1'
os.environ['TF_ADJUST_HUE_FUSED'] = '1'
os.environ['TF_ADJUST_SATURATION_FUSED'] = '1'
os.environ['TF_ENABLE_WINOGRAD_NONFUSED'] = '1'
os.environ['TF_SYNC_ON_FINISH'] = '0'
os.environ['TF_AUTOTUNE_THRESHOLD'] = '2'
os.environ['TF_DISABLE_NVTX_RANGES'] = '1'
if params['use_amp']:
assert params['dtype'] == tf.float32, "TF-AMP requires FP32 precision"
LOGGER.log("TF AMP is activated - Experimental Feature")
os.environ['TF_ENABLE_AUTO_MIXED_PRECISION'] = '1'
runner = Runner(params)
if 'train' in params['exec_mode'] \
or 'train_and predict' in params['exec_mode']:
runner.train()
if 'train_and predict' in params['exec_mode'] \
or 'predict' in params['exec_mode']:
runner.predict()
if 'benchmark' in params['exec_mode']:
runner.benchmark()
