ai-operator-standard

优化器算子接口

框架版本：

• Tensorflow：2.9.0
• PyTorch：1.11.0

算子	TensorFlow	PyTorch	MindSpore	Paddle
sgd	tf.keras.optimizers.SGD	torch.optim.SGD
momentum	tf.keras.optimizers.SGD	torch.optim.SGD
adagrad	tf.keras.optimizers.Adagrad	torch.optim.Adagrad
adadelta	tf.keras.optimizers.Adadelta	torch.optim.Adadelta
rmsprop	tf.keras.optimizers.RMSprop	torch.optim.RMSprop
centeredrmsprop	tf.raw_ops.ApplyCenteredRMSProp	/
adam	tf.keras.optimizers.Adam	torch.optim.Adam
adamax	tf.keras.optimizers.Adamax	torch.optim.Adamax

---

图像处理算子接口

框架版本：

• Tensorflow：2.9.0
• PyTorch：1.11.0
• torchvision：0.12

算子	TensorFlow	PyTorch	MindSpore	Paddle
adjust_brightness	tf.image.adjust_brightness	torchvision.transforms.functional.adjust_brightness
adjust_contrast	tf.image.adjust_contrast	torchvision.transforms.functional.adjust_contrast
adjust_gamma	tf.image.adjust_gamma	torchvision.transforms.functional.adjust_gamma
adjust_hue	tf.image.adjust_hue	torchvision.transforms.functional.adjust_hue
adjust_saturation	tf.image.adjust_saturation	torchvision.transforms.functional.adjust_saturation
random_brightness	tf.image.random_brightness	/
random_contrast	tf.image.random_contrast	/
random_hue	tf.image.random_hue	/
random_saturation	tf.image.random_saturation	/
rgb_to_grayscale	tf.image.rgb_to_grayscale	torchvision.transforms.functional.rgb_to_grayscale
grayscale_to_rgb	tf.image.grayscale_to_rgb	/
rgb_to_hsv	tf.image.rgb_to_hsv	/
hsv_to_rgb	tf.image.hsv_to_rgb	/
rgb_to_yuv	tf.image.rgb_to_yuv	/
yuv_to_rgb	tf.image.yuv_to_rgb	/
rgb_to_yiq	tf.image.rgb_to_yiq	/
yiq_to_rgb	tf.image.yiq_to_rgb	/
convert_image_dtype	tf.image.convert_image_dtype	torchvision.transforms.functional.convert_image_dtype
horizontal_flip	tf.image.flip_left_right	torchvision.transforms.functional.hflip
vertical_flip	tf.image.flip_up_down	torchvision.transforms.functional.vflip
random_horizontal_flip	tf.image.random_flip_left_right	torchvision.transforms.RandomHorizontalFlip
random_vertical_flip	tf.image.random_flip_up_down	torchvision.transforms.RandomVerticalFlip
image_transpose	tf.image.transpose	torch.transpose
rot90	tf.image.rot90	torch.rot90
center_crop	tf.image.central_crop	torchvision.transforms.CenterCrop
random_crop	tf.image.random_crop	torchvision.transforms.RandomCrop
crop_to_bounding_box	tf.image.crop_to_bounding_box	/
pad_to_bounding_box	tf.image.pad_to_bounding_box	torchvision.transforms.Pad
resize_with_crop_or_pad	tf.image.resize_with_crop_or_pad	torchvision.transforms.functional.resized_crop
resize	tf.image.resize	torchvision.transforms.functional.resize
Gaussian_blur	/	torchvision.transforms.functional.gaussian_blur
image_gradients	tf.image.image_gradients	/
image_normalize	/	torchvision.transforms.functional.normalize

---

分布式训练算子接口

框架版本：

• Tensorflow：2.9.0
• PyTorch：1.11.0

算子	TensorFlow	PyTorch	MindSpore	Paddle
comm_init	tf.raw_ops.CollectiveInitializeCommunicator	torch.distributed.init_process_group
comm_create_by_ranks	/	torch.distributed.new_group
comm_dup	/	/
comm_destroy	/	/
comm_get_size	/	torch.distributed.get_world_size
comm_get_rank	/	torch.distributed.get_rank
get_async_op_status	/	is_completed
wait	/	wait
comm_send	tf.raw_ops.Send	torch.distributed.send
comm_recv	tf.raw_ops.Recv	torch.distributed.recv
comm_broadcast	tf.raw_ops.NcclBroadcast	torch.distributed.broadcast
comm_all_reduce	tf.distribute.NcclAllReduce	torch.distributed.all_reduce
comm_reduce	tf.raw_ops.NcclReduce	torch.distributed.reduce
comm_all_gather	/	torch.distributed.all_gather
comm_gather	tf.raw_ops.CollectiveGatherV2	torch.distributed.gather
comm_scatter	/	torch.distributed.scatter
comm_reduce_scatter	/	torch.distributed.reduce_scatter
comm_all_to_all	tf.raw_ops.AllToAll	torch.distributed.all_to_all
comm_barrier	/	torch.distributed.barrier
comm_send_async	/	torch.distributed.isend
comm_recv_async	/	torch.distributed.irecv
comm_broadcast_async	/	torch.distributed.broadcast
comm_all_reduce_async	/	torch.distributed.all_reduce
comm_reduce_async	/	torch.distributed.reduce
comm_all_gather_async	/	torch.distributed.all_gather
comm_gather_async	/	torch.distributed.gather
comm_scatter_async	/	torch.distributed.scatter
comm_reduce_scatter_async	/	torch.distributed.reduce_scatter
comm_all_to_all_async	/	torch.distributed.all_to_all
comm_barrier_async	/	torch.distributed.barrier