def get_backend(backend):
if backend == "tensorflow":
from backend_tf import BackendTensorflow
backend = BackendTensorflow()
elif backend == "onnxruntime":
from backend_onnxruntime import BackendOnnxruntime
backend = BackendOnnxruntime()
elif backend == "null":
from backend_null import BackendNull
backend = BackendNull()
elif backend == "pytorch":
from backend_pytorch import BackendPytorch
backend = BackendPytorch()
elif backend == "pytorch-native":
from backend_pytorch_native import BackendPytorchNative
backend = BackendPytorchNative()
elif backend == "tflite":
from backend_tflite import BackendTflite
backend = BackendTflite()
elif backend == "tflite-calibrate":
from backend_tflite_calibrate import BackendTflite
backend = BackendTflite()
elif backend == "tflite-ncore":
from backend_tflite_ncore import BackendTfliteNcore
backend = BackendTfliteNcore()
elif backend == "tflite-ncore-offline-imagenet":
from backend_tflite_ncore_offline_imagenet import BackendTfliteNcoreOfflineImagenet
backend = BackendTfliteNcoreOfflineImagenet()
elif backend == "tflite-ncore-offline-ssd":
from backend_tflite_ncore_offline_ssd import BackendTfliteNcoreOfflineSSD
backend = BackendTfliteNcoreOfflineSSD()
else:
raise ValueError("unknown backend: " + backend)
return backend
def get_backend(backend):
if backend == "tensorflow":
from backend_tf import BackendTensorflow
backend = BackendTensorflow()
elif backend == "onnxruntime":
from backend_onnxruntime import BackendOnnxruntime
backend = BackendOnnxruntime()
elif backend == "null":
from backend_null import BackendNull
backend = BackendNull()
elif backend == "pytorch":
from backend_pytorch import BackendPytorch
backend = BackendPytorch()
elif backend == "pytorch-native":
from backend_pytorch_native import BackendPytorchNative
backend = BackendPytorchNative()
elif backend == "tflite":
from backend_tflite import BackendTflite
backend = BackendTflite()
elif backend == "tvm":
from backend_tvm import BackendTvm
backend = BackendTvm()
else:
raise ValueError("unknown backend: " + backend)
return backend
def get_backend(backend):
if backend == "tensorflow":
from backend_tf import BackendTensorflow
backend = BackendTensorflow()
elif backend == "onnxruntime":
from backend_onnxruntime import BackendOnnxruntime
backend = BackendOnnxruntime()
elif backend == "null":
from backend_null import BackendNull
backend = BackendNull()
elif backend == "pytorch":
from backend_pytorch import BackendPytorch
backend = BackendPytorch()
elif backend == "pytorch-native":
from backend_pytorch_native import BackendPytorchNative
backend = BackendPytorchNative()
elif backend == "pytorch-centaur":
from backend_pytorch_centaur import BackendPytorchCentaur
backend = BackendPytorchCentaur()
elif backend == "pytorch-native-calibrate":
from backend_pytorch_native_calibrate import BackendPytorchNativeCalibrate
backend = BackendPytorchNativeCalibrate()
elif backend == "tflite":
from backend_tflite import BackendTflite
backend = BackendTflite()
elif backend == "tflite-calibrate":
from backend_tflite_calibrate import BackendTflite
backend = BackendTflite()
elif backend == "tflite-ncore":
from backend_tflite_ncore import BackendTfliteNcore
backend = BackendTfliteNcore()
elif backend == "tflite-ncore-mobilenet":
from backend_libncoretflite import BackendTfliteNcoreMobileNetV1
backend = BackendTfliteNcoreMobileNetV1()
backend.inputs = ["image_tensor:0"]
elif backend == "tflite-ncore-resnet":
from backend_libncoretflite import BackendTfliteNcoreResnet
backend = BackendTfliteNcoreResnet()
backend.inputs = ["image_tensor:0"]
elif backend == "tflite-ncore-ssd":
from backend_libncoretflite import BackendTfliteNcoreSSD
backend = BackendTfliteNcoreSSD()
backend.inputs = ["image_tensor:0"]
elif backend == "tflite-ncore-mobilenet-offline":
from backend_libncoretflite import BackendTfliteNcoreMobileNetV1Offline
backend = BackendTfliteNcoreMobileNetV1Offline()
backend.inputs = ["image_tensor:0"]
elif backend == "tflite-ncore-resnet-offline":
from backend_libncoretflite import BackendTfliteNcoreResnetOffline
backend = BackendTfliteNcoreResnetOffline()
backend.inputs = ["image_tensor:0"]
elif backend == "tflite-ncore-ssd-offline":
from backend_libncoretflite import BackendTfliteNcoreSSDOffline
backend = BackendTfliteNcoreSSDOffline()
backend.inputs = ["image_tensor:0"]
else:
raise ValueError("unknown backend: " + backend)
return backend
def get_backend(backend, dataset_path, dataset_calibration_list):
if backend == "tensorflow":
from backend_tf import BackendTensorflow
backend = BackendTensorflow()
elif backend == "onnxruntime":
from backend_onnxruntime import BackendOnnxruntime
backend = BackendOnnxruntime()
elif backend == "null":
from backend_null import BackendNull
backend = BackendNull()
elif backend == "pytorch":
from backend_pytorch import BackendPytorch
backend = BackendPytorch()
elif backend == "pytorch-native":
from backend_pytorch_native import BackendPytorchNative
backend = BackendPytorchNative()
elif backend == "pytorch-jit-traced":
from backend_pytorch_jit_traced import BackendPytorchJITTraced
backend = BackendPytorchJITTraced()
elif backend == "pytorch-fp32":
from backend_pytorch_fp32 import BackendPytorchFP32
backend = BackendPytorchFP32()
elif backend == "pytorch-ssd-jit-traced":
from backend_pytorch_ssd_jit_traced import BackendPytorchSSDJITTraced
backend = BackendPytorchSSDJITTraced()
elif backend == "pytorch-yolov3-jit-traced":
from backend_pytorch_yolov3_jit_traced import BackendPytorchYOLOv3JITTraced
backend = BackendPytorchYOLOv3JITTraced()
elif backend == "pytorch-yolov3-fp32":
from backend_pytorch_yolov3_fp32 import BackendPytorchYOLOv3FP32
backend = BackendPytorchYOLOv3FP32()
elif backend == "tflite":
from backend_tflite import BackendTflite
backend = BackendTflite()
elif backend == "edgecortix":
from backend_edgecortix import BackendEdgecortix
backend = BackendEdgecortix(dataset_path, dataset_calibration_list)
else:
raise ValueError("unknown backend: " + backend)
return backend
def get_backend(backend, dataset, max_ind_range, data_sub_sample_rate, use_gpu):
if backend == "pytorch-native":
from backend_pytorch_native import BackendPytorchNative
# NOTE: pass model parameters here, the following options are available
if dataset == "kaggle":
# 1. Criteo Kaggle Display Advertisement Challenge Dataset (see ./bench/dlrm_s_criteo_kaggle.sh)
backend = BackendPytorchNative(
m_spa=16,
ln_emb=np.array([1460,583,10131227,2202608,305,24,12517,633,3,93145,5683,8351593,3194,27,14992,5461306,10,5652,2173,4,7046547,18,15,286181,105,142572]),
ln_bot=np.array([13,512,256,64,16]),
ln_top=np.array([367,512,256,1]),
use_gpu=use_gpu
)
elif dataset == "terabyte":
if max_ind_range == 10000000:
# 2. Criteo Terabyte (see ./bench/dlrm_s_criteo_terabyte.sh [--sub-sample=0.875] --max-in-range=10000000)
backend = BackendPytorchNative(
m_spa=64,
ln_emb=np.array([9980333,36084,17217,7378,20134,3,7112,1442,61, 9758201,1333352,313829,10,2208,11156,122,4,970,14, 9994222, 7267859, 9946608,415421,12420,101, 36]),
ln_bot=np.array([13,512,256,64]),
ln_top=np.array([415,512,512,256,1]),
use_gpu=use_gpu
)
elif max_ind_range == 40000000:
# 3. Criteo Terabyte MLPerf training (see ./bench/run_and_time.sh --max-in-range=40000000)
backend = BackendPytorchNative(
m_spa=128,
ln_emb=np.array([39884406,39043,17289,7420,20263,3,7120,1543,63,38532951,2953546,403346,10,2208,11938,155,4,976,14,39979771,25641295,39664984,585935,12972,108,36]),
ln_bot=np.array([13,512,256,128]),
ln_top=np.array([479,1024,1024,512,256,1]),
use_gpu=use_gpu
)
else:
raise ValueError("only --max-in-range 10M or 40M is supported")
else:
raise ValueError("only kaggle|terabyte dataset options are supported")
elif backend == "onnxruntime":
from backend_onnxruntime import BackendOnnxruntime
# NOTE: pass model parameters here, the following options are available
if dataset == "kaggle":
# 1. Criteo Kaggle Display Advertisement Challenge Dataset (see ./bench/dlrm_s_criteo_kaggle.sh)
backend = BackendOnnxruntime(
m_spa=16,
ln_emb=np.array([1460,583,10131227,2202608,305,24,12517,633,3,93145,5683,8351593,3194,27,14992,5461306,10,5652,2173,4,7046547,18,15,286181,105,142572]),
ln_bot=np.array([13,512,256,64,16]),
ln_top=np.array([367,512,256,1]),
use_gpu=use_gpu
)
elif dataset == "terabyte":
if max_ind_range == 10000000:
# 2. Criteo Terabyte (see ./bench/dlrm_s_criteo_terabyte.sh [--sub-sample=0.875] --max-in-range=10000000)
backend = BackendOnnxruntime(
m_spa=64,
ln_emb=np.array([9980333,36084,17217,7378,20134,3,7112,1442,61, 9758201,1333352,313829,10,2208,11156,122,4,970,14, 9994222, 7267859, 9946608,415421,12420,101, 36]),
ln_bot=np.array([13,512,256,64]),
ln_top=np.array([415,512,512,256,1]),
use_gpu=use_gpu
)
elif max_ind_range == 40000000:
# 3. Criteo Terabyte MLPerf training (see ./bench/run_and_time.sh --max-in-range=40000000)
backend = BackendOnnxruntime(
m_spa=128,
ln_emb=np.array([39884406,39043,17289,7420,20263,3,7120,1543,63,38532951,2953546,403346,10,2208,11938,155,4,976,14,39979771,25641295,39664984,585935,12972,108,36]),
ln_bot=np.array([13,512,256,128]),
ln_top=np.array([479,1024,1024,512,256,1]),
use_gpu=use_gpu
)
else:
raise ValueError("only --max-in-range 10M or 40M is supported")
else:
raise ValueError("only kaggle|terabyte dataset options are supported")
else:
raise ValueError("unknown backend: " + backend)
return backend
