Function-Level Support Status

ONNX Support Status

Note: In this document, the numbers in the header of all tables represent the version of onnx opset.

Import

✓: onnx specification defined, and supported.
X: onnx specification defined, but not support yet.
Empty: Not defined (Support status follows latest).

Total: 93/155

ONNX Operator

1

2

3

4

5

6

7

8

9

10

11

NNabla Func

Description

Abs Acos Acosh Add And ArgMax ArgMin Asin Asinh Atan Atanh AveragePool BatchNormalization BitShift Cast Ceil Clip Compress Concat ConcatFromSequence Constant ConstantOfShape Conv ConvInteger ConvTranspose Cos Cosh CumSum DepthToSpace DequantizeLinear Det Div Dropout DynamicQuantizeLinear Elu Equal Erf Exp Expand EyeLike Flatten Floor GRU Gather GatherElements GatherND Gemm GlobalAveragePool GlobalLpPool GlobalMaxPool Greater HardSigmoid Hardmax Identity If InstanceNormalization IsInf IsNaN LRN LSTM LeakyRelu Less Log LogSoftmax Loop LpNormalization LpPool MatMul MatMulInteger Max MaxPool MaxRoiPool MaxUnpool Mean MeanVarianceNormalization Min Mod Mul Multinomial Neg NonMaxSuppression NonZero Not OneHot Or PRelu Pad Pow QLinearConv QLinearMatMul QuantizeLinear RNN RandomNormal RandomNormalLike RandomUniform RandomUniformLike Range Reciprocal ReduceL1 ReduceL2 ReduceLogSum ReduceLogSumExp ReduceMax ReduceMean ReduceMin ReduceProd ReduceSum ReduceSumSquare Relu Reshape Resize ReverseSequence RoiAlign Round Scan Scatter ScatterElements ScatterND Selu SequenceAt SequenceConstruct SequenceErase SequenceInsert SequenceLength Shape Shrink Sigmoid Sign Sin Sinh Size Slice Softmax Softplus Softsign SpaceToDepth Split SplitToSequence Sqrt Squeeze StringNormalizer Sub Sum Tan Tanh TfIdfVectorizer ThresholdedRelu Tile TopK Transpose Unique Unsqueeze Upsample Where Xor

✓

✓ ✓ ✓ ✓

✓ X

✓ ✓ ✓

✓

✓ ✓

✓

✓ ✓ X ✓

✓ ✓ X X ✓ ✓ ✓ ✓ X ✓

✓ X ✓ ✓ ✓ ✓ X X X ✓

✓ ✓ X

✓

✓ ✓ ✓ ✓

X X X X X

✓ X X X X ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓

✓

X

✓

X ✓ ✓ ✓ ✓ ✓ ✓

✓ ✓

✓

✓ X ✓

✓ X

✓

X

✓

X

✓

X

✓ ✓ ✓

✓ ✓

✓

✓ ✓

✓

✓ ✓

✓ ✓ ✓ ✓ ✓

✓

✓ X

✓ ✓ ✓

✓

✓ ✓

✓

✓ ✓

✓

✓ ✓ X ✓

✓ ✓

✓ ✓ ✓ ✓

✓

✓ X ✓ ✓ ✓ ✓

✓

✓ ✓

✓

✓ X ✓

✓

✓ ✓ ✓ ✓

X

✓

✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓

✓

✓ ✓ ✓ ✓ ✓ ✓

✓ ✓

✓ ✓ ✓ ✓

✓

✓ X

✓

✓ X

✓

X

✓

X

X ✓

✓

X

✓ X ✓

✓

X ✓ X ✓

X

X X

X

✓

X

✓ ✓

X X

X

✓

X

X X X

✓

X

✓

X

✓ ✓

X

✓ X X X X

X

X ✓

X

✓

✓ X X ✓

✓

✓ X

X

✓

X

X X X X ✓ ✓ ✓ ✓ ✓ ✓

X

✓ X X X X

X X X X X

X ✓

✓ X

✓

X

X ✓

Abs ACos ACosh Reshape, Add2 Reshape, LogicalAnd Max Min ASin ASinh ATan ATanh Pad, AveragePooling BatchNormalization

Abs, Log Ceil MinimumScalar, Identity, MaximumScalar

Concatenate

Identity Constant Convolution

Deconvolution, Pad Cos Cosh

Reshape, Transpose

Reshape, Div2 Identity

ELU Equal, Reshape

Exp Reshape, Broadcast

Reshape Floor

Slice, Concatenate

MulScalar, Reshape, Add2, BatchMatmul GlobalAveragePooling

Reshape, Greater MulScalar, AddScalar, HardSigmoid, MinimumScalar, MaximumScalar Max, Reshape, OneHot Identity

BatchNormalization, Reshape, Concatenate, Split IsInf IsNaN SumPooling, Transpose, MulScalar, AddScalar, PowScalar, Div2

LeakyReLU Reshape, Less Log Add2, Sum, Exp, Sub2, Reshape, Max, Log

Reshape, BatchMatmul

Maximum2 MaxPooling, Pad

Stack, Broadcast, Mean

Minimum2

Reshape, Mul2

MulScalar

LogicalNot

Reshape, LogicalOr PReLU Pad Reshape, Pow2

RDivScalar

Max Mean Min Prod Sum PowScalar, Sum ReLU Reshape

Round

SELU

Sigmoid Sign Sin Sinh

Slice Sum, Exp, Sub2, Reshape, Max, Div2 SoftPlus SoftSign Reshape, Transpose Stack, Split

PowScalar Reshape

Sub2, Reshape AddN Tan Tanh

GreaterScalar, Constant, Where Tile

Transpose

Reshape Unpooling Where Reshape, LogicalXor

Not all features are verified. Those features can be verified by ONNXRuntime when opset > 6. Some feature is not supported by Nnabla such as Pad's edge mode. if opset >= 10, the ceil_mode is not supported.

Not yet implemented.

Not yet implemented. Not yet implemented.

Not yet implemented.

Not yet implemented. Not yet implemented.

Not yet implemented.

Not yet implemented. Not yet implemented. Not yet implemented.

Not yet implemented.

Not all features are verified. Those features can be verified by ONNXRuntime. if opset >= 10, the ceil_mode is not supported, dilations is not equal to 1 is not supported. Not yet implemented. Not yet implemented.

Not yet implemented.

Not yet implemented. Not yet implemented.

Not yet implemented.

Onnx required to support "edge" mode, while nnabla does not support it.

Not yet implemented. Not yet implemented. Not yet implemented. Not yet implemented. Not yet implemented. Not yet implemented. Not yet implemented. Not yet implemented. Not yet implemented.

Not yet implemented. Not yet implemented. Not yet implemented. Not yet implemented.

Not yet implemented. Not yet implemented. Not yet implemented.

Not yet implemented. Not yet implemented. Not yet implemented. Not yet implemented.

Not yet implemented. Not yet implemented. Not yet implemented. Not yet implemented. Not yet implemented. Not yet implemented. Not yet implemented.

Not yet implemented.

Export

✓: Support to export this opset.
△: Partially support to export this opset (e.g. some cases cannot be supported, or not completely tested).
X: Supported, but test failed.
Empty: Not support corresponding opset version.

Total: 120/173

Neural Network Layer

Count 11/14

NNabla Function 7 9 10 11 ONNX Op Description

Affine RNN LSTM GRU Convolution DepthwiseConvolution Deconvolution DepthwiseDeconvolution MaxPooling

✓

✓ ✓ ✓ ✓ ✓

✓

✓ ✓ ✓ ✓ ✓

✓

✓ ✓ ✓ ✓ ✓

✓

✓ ✓ ✓ ✓ ✓

Reshape, Gemm

Reshape, Conv Reshape, Conv ConvTranspose, Reshape ConvTranspose, Reshape Constant, Reshape, Pad, MaxPool

Not yet implemented. Not yet implemented. Not yet implemented.

AveragePooling GlobalAveragePooling SumPooling Unpooling Embed

△ ✓ ✓ ✓ ✓

△ ✓ ✓ ✓ ✓

△ ✓ ✓ ✓ ✓

△ ✓ ✓ ✓ ✓

AveragePool, Constant, Reshape, Pad GlobalAveragePool Constant, Mul, Pad, AveragePool, Reshape Resize Gather

Currently only supports the cases where both ignore_border and including_pad are True.

Neural Network Activation Functions

Count 21/21

NNabla Function 7 9 10 11 ONNX Op Description

Sigmoid Swish Tanh ReLU LeakyReLU Softmax LogSoftmax ELU SELU CReLU CELU PReLU GELU ReLU6 HardSigmoid HardTanh LogSigmoid SoftPlus SoftSign TanhShrink Sinc

✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ X

✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ X

✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ X

✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓

Sigmoid Mul, Sigmoid Tanh Relu LeakyRelu Sub, Div, Exp, ReduceSum, ReduceMax Sub, Exp, ReduceSum, ReduceMax, Log Elu Selu Concat, Neg, Relu Concat, Neg, Elu Reshape, PRelu Constant, Mul, Add, Tanh, Div, Pow, Sqrt Min, Constant, Relu HardSigmoid Max, Neg, Constant, Min Sigmoid, Log Softplus Softsign Sub, Tanh Constant, Where, Sin, Div, Equal

Normalization

Count 2/6

NNabla Function 7 9 10 11 ONNX Op Description

FusedBatchNormalization BatchNormalization SyncBatchNormalization MeanSubtraction ClipGradByValue ClipGradByNorm

✓ ✓

✓ ✓

✓ ✓

✓ ✓

Squeeze, Sub, Constant, Mul, Add, Relu, Div, BatchNormalization, Reshape, ReduceSum, ReduceMean Squeeze, Sub, Constant, Mul, Div, BatchNormalization, Reshape, ReduceSum, ReduceMean

Not yet implemented. Not yet implemented. Not yet implemented. Not yet implemented.

Reduction

Count 5/7

NNabla Function 7 9 10 11 ONNX Op Description

Sum Mean Max Min Prod ReduceSum ReduceMean

✓ ✓ ✓ ✓ ✓

✓ ✓ ✓ ✓ ✓

✓ ✓ ✓ ✓ ✓

✓ ✓ ✓ ✓ ✓

ReduceSum ReduceMean ReduceMax ReduceMin ReduceProd

Not yet implemented. Not yet implemented.

Arithmetic

Count 11/12

NNabla Function 7 9 10 11 ONNX Op Description

Add2 BcAdd2 Sub2 Mul2 Div2 Pow2 AddScalar MulScalar PowScalar RSubScalar RDivScalar RPowScalar

✓

✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓

✓

✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓

✓

✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓

✓

✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓

Add

Sub Mul Div Pow Add, Constant Constant, Mul Pow, Constant Sub, Constant Constant, Div Pow, Constant

Not yet implemented.

Logical

Count 29/29

NNabla Function 7 9 10 11 ONNX Op Description

Sign Minimum2 Maximum2 MinimumScalar MaximumScalar LogicalAnd LogicalOr LogicalXor Equal NotEqual GreaterEqual Greater LessEqual Less LogicalAndScalar LogicalOrScalar LogicalXorScalar EqualScalar NotEqualScalar GreaterEqualScalar GreaterScalar LessEqualScalar LessScalar LogicalNot IsNaN IsInf ResetNaN ResetInf Where

X ✓ ✓ ✓ ✓ ✓ ✓ ✓ X X ✓ ✓ ✓ ✓ ✓ ✓ ✓ X X ✓ ✓ ✓ ✓ ✓ X X X X X

✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ X X ✓ ✓ ✓ ✓ ✓ ✓ ✓ X X ✓ ✓ ✓ ✓ ✓ ✓ X ✓ X ✓

✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ X X ✓ ✓ ✓ ✓ ✓ ✓ ✓ X X ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓

✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓

Sign Add, Constant, Min Max, Add, Constant Add, Constant, Min Max, Add, Constant And Or Xor Equal Not, Equal Not, Less Greater Not, Greater Less Constant, And Constant, Or Constant, Xor Constant, Equal Not, Equal, Constant Not, Constant, Less Constant, Greater Not, Constant, Greater Constant, Less Not IsNaN IsInf Constant, Where, IsNaN IsInf, Constant, Where Where

Math

Count 22/22

NNabla Function 7 9 10 11 ONNX Op Description

Constant Arange Abs Exp Log Identity BatchMatmul Round Ceil Floor Sin Cos Tan Sinh Cosh ASin ACos ATan ATan2 ASinh ACosh ATanh

✓ ✓ ✓ ✓ ✓ ✓ ✓ X ✓ ✓ ✓ ✓ ✓ X X ✓ ✓ ✓ ✓ X X X

✓ ✓ ✓ ✓ ✓ ✓ ✓ X ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓

✓ ✓ ✓ ✓ ✓ ✓ ✓ X ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓

✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓

Constant, Identity Constant, Identity Abs Exp Log Identity Transpose, MatMul Round Ceil Floor Sin Cos Tan Sinh Cosh Asin Acos Atan Atan, Div Asinh Acosh Atanh

Array Manipulation

Count 12/19

NNabla Function 7 9 10 11 ONNX Op Description

Concatenate Split Stack

✓ ✓ ✓

✓ ✓ ✓

✓ ✓ ✓

✓ ✓ ✓

Concat Squeeze, Split Concat, Unsqueeze

Slice

△ △ ✓ ✓ Slice, Constant ONNX slice cannot support step != 1 on opset < 10.

Pad Transpose Broadcast BroadcastTo Tile OneHot Flip Shift Sort Reshape MatrixDiag MatrixDiagPart Assign GatherNd ScatterNd

△ ✓ X ✓ ✓ X ✓

✓

△ ✓ ✓ ✓ ✓ ✓ ✓

✓

△ ✓ ✓ ✓ ✓ ✓ ✓

✓

△ ✓ ✓ ✓ ✓ ✓ ✓

✓

Constant, Pad Transpose

Tile, Constant, Reshape Flatten, Reshape, Gather Identity, Transpose, Gather

Constant, Reshape

When the mode of the pad is reflect, if the size of the pad exceeds the input size, onnxruntime cannot handle it.

Not yet implemented. Not yet implemented.

Not yet implemented. Not yet implemented. Not yet implemented. Not yet implemented. Not yet implemented.

Signal Processing

Count 1/3

NNabla Function 7 9 10 11 ONNX Op Description

Interpolate FFT IFFT

X

X

△

✓

Resize, Reshape

Not yet implemented. Not yet implemented.

Stochasticity

Count 0/11

NNabla Function 7 9 10 11 ONNX Op Description

Dropout TopKData TopKGrad Rand Randint Randn RandomChoice RandomCrop RandomFlip RandomShift ImageAugmentation

X

X

X

X

Dropout

The Dropout in nnabla has no test mode and contains random parameters, so the test result is not the same as onnx. Not yet implemented. Not yet implemented. Not yet implemented. Not yet implemented. Not yet implemented. Not yet implemented. Not yet implemented. Not yet implemented. Not yet implemented. Not yet implemented.

Loss Functions

Count 0/9

NNabla Function 7 9 10 11 ONNX Op Description

SigmoidCrossEntropy BinaryCrossEntropy SoftmaxCrossEntropy CategoricalCrossEntropy SquaredError AbsoluteError HuberLoss EpsilonInsensitiveLoss KLMultinomial

Not yet implemented. Not yet implemented. Not yet implemented. Not yet implemented. Not yet implemented. Not yet implemented. Not yet implemented. Not yet implemented. Not yet implemented.

Quantization Neural Network Layers

Count 6/12

NNabla Function 7 9 10 11 ONNX Op Description

BinarySigmoid BinaryTanh BinaryConnectAffine BinaryConnectConvolution BinaryWeightAffine BinaryWeightConvolution INQAffine INQConvolution FixedPointQuantize MinMaxQuantize Pow2Quantize Prune

X X ✓ ✓ ✓ ✓

✓ ✓ ✓ ✓ ✓ ✓

✓ ✓ ✓ ✓ ✓ ✓

✓ ✓ ✓ ✓ ✓ ✓

Constant, Greater, Where Constant, Greater, Where Reshape, Gemm Reshape, Conv Add, Reshape, Mul, MatMul Add, Mul, Reshape, Conv

Not yet implemented. Not yet implemented. Not yet implemented. Not yet implemented. Not yet implemented. Not yet implemented.

Validation

Count 0/3

NNabla Function 7 9 10 11 ONNX Op Description

TopNError BinaryError ConfusionMatrix

Not yet implemented. Not yet implemented. Not yet implemented.

Unsupported, Special Use

Count 0/5

NNabla Function 7 9 10 11 ONNX Op Description

VATNoise Unlink Sink NmsDetection2d MaxPoolingBackward

Not yet implemented. Not yet implemented. Not yet implemented. Not yet implemented. Not yet implemented.

Tensorflow Support Status

Import

✓: Supported
△: Partially supported
X: Supported, but test failed.
Empty: Not support yet.

Total: 109/122

Tensorflow support status

Tensorflow Function

Status

NNabla Func

Description

Abs Acos Acosh Add AddN All Any ArgMax ArgMin Asin Asinh Atan Atan2 Atanh AvgPool AvgPool3D BatchMatMul BatchNormalization BiasAdd BroadcastTo

✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ △ △ ✓ ✓ ✓ ✓

Abs ACos ACosh Add2 AddN Reshape, Min, Greater Reshape, Sum, Greater Max Min ASin ASinh ATan Add2, Sub2, Reshape, Mul2, Sign, ATan, Div2 ATanh Transpose, Pad, AveragePooling AveragePooling, Transpose, Pad Transpose, BatchMatmul RDivScalar, Add2, Sub2, Reshape, Mul2, PowScalar Reshape, Add2

Cast Ceil ClipByValue Concat ConcatV2 Const Conv1D Conv1DTranspose Conv2D Conv2DBackpropInput Conv3D Conv3DBackpropInput Cos Cosh Crelu Cumsum DepthToSpace DepthwiseConv2d Div Elu Equal Erf Erfc Exp ExpandDims Floor FloorDiv FloorMod GatherNd

X ✓ ✓ ✓ ✓ ✓ △ △ △ △ △ △ ✓ ✓ ✓ X ✓ △ ✓ ✓ ✓ X X ✓ ✓ ✓ ✓ ✓ X

NA Ceil Maximum2, Reshape, Minimum2 Concatenate Concatenate NA Convolution, Reshape, Transpose, Pad Deconvolution, Reshape, Transpose Convolution, Transpose, Pad Deconvolution, Transpose Convolution, Transpose, Pad Deconvolution, Transpose, Pad Cos Cosh ReLU, MulScalar, Concatenate

Reshape, Transpose Convolution, Reshape, Transpose, Pad Div2 ELU Equal

Exp Reshape Floor Floor, Div2 Sub2, Floor, Div2, Mul2

Not yet implemented.

Not yet implemented. Not yet implemented.

Not yet implemented.

GatherV2 Greater GreaterEqual Identity IsInf IsNan LeakyRelu Less LessEqual Log LogSigmoid LogSoftmax LogicalAnd LogicalNot LogicalOr LogicalXor Max MaxPool MaxPool3D MaxPoolWithArgmax Maximum Mean Min Minimum Mul Neg NotEqual Pack Pad Pow Prod RealDiv Reciprocal Relu Relu6 Reshape ReverseSequence ReverseV2 Round Rsqrt Selu Shape Sigmoid Sign Sin Sinh Size Slice Softmax Softplus Softsign SpaceToDepth Split SplitV Sqrt Square SquaredDifference Squeeze StopGradient StridedSlice Sub Sum Swish Tan Tanh Tile TopKV2 Transpose TruncateDiv TruncateMod Unpack Where ZerosLike

X ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ △ △ X ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ △ ✓ ✓ ✓ ✓ ✓ ✓ ✓ X X ✓ ✓ ✓ X ✓ ✓ ✓ ✓ X ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ △ ✓ ✓ ✓ ✓ ✓ ✓ X ✓ ✓ X ✓ △ ✓

Slice, Concatenate Greater Less, LogicalNot Identity IsInf IsNaN LeakyReLU Less LogicalNot, Greater Log SoftPlus, MulScalar Add2, Sum, Transpose, Exp, Sub2, Reshape, Max, Log LogicalAnd LogicalNot LogicalOr LogicalAnd, LogicalNot, LogicalOr Max Reshape, Transpose, MaxPooling, Pad MaxPooling, Pad, Transpose

Maximum2 Mean Min Minimum2 Mul2 MulScalar Equal, LogicalNot Reshape, Concatenate Pad Pow2 Prod Div2 RDivScalar ReLU MinimumScalar, MaximumScalar Reshape

Round RDivScalar, PowScalar SELU

Sigmoid Sign Sin Sinh

Slice Transpose, Sum, Exp, Sub2, Reshape, Max, Div2 SoftPlus SoftSign Reshape, Transpose Stack, Split Stack, Split PowScalar Mul2 Sub2, Mul2 Reshape Identity Slice Sub2 Sum Sigmoid, Mul2 Tan Tanh Tile

Transpose Div2

Stack, Reshape, Split Where NA

Not yet implemented.

Not yet implemented. Not yet implemented.

Not yet implemented.

Export

✓: Supported
△: Partially supported
X: Supported, but test failed.
Empty: Not support yet.

Total: 120/173

Neural Network Layer

Count 11/14

NNabla Function Status Description

Affine RNN LSTM GRU

✓

Not yet implemented. Not yet implemented. Not yet implemented.

Convolution

△ The cases dilations and strides larger than 1 are not supported by tensorflow.

DepthwiseConvolution

△ The cases dilations and strides larger than 1 are not supported by tensorflow.

Deconvolution

△ The cases dilations larger than 1 are not supported by tensorflow.

DepthwiseDeconvolution MaxPooling

△ ✓

The cases dilations larger than 1 are not supported by tensorflow.

AveragePooling GlobalAveragePooling SumPooling

△ ✓ ✓

Currently only supports the cases both ignore_border and including_pad are True.

Unpooling Embed

△ ✓

The kernel only supports 2d.

Neural Network Activation Functions

Count 21/21

NNabla Function Status Description

Sigmoid Swish Tanh ReLU LeakyReLU Softmax LogSoftmax ELU SELU CReLU CELU PReLU GELU ReLU6 HardSigmoid HardTanh LogSigmoid SoftPlus SoftSign TanhShrink Sinc

✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ △ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓

Normalization

Count 2/6

NNabla Function Status Description

FusedBatchNormalization BatchNormalization SyncBatchNormalization MeanSubtraction ClipGradByValue ClipGradByNorm

✓ ✓

Not yet implemented. Not yet implemented. Not yet implemented. Not yet implemented.

Reduction

Count 5/7

NNabla Function Status Description

Sum Mean Max Min Prod ReduceSum ReduceMean

✓ ✓ ✓ ✓ ✓

Not yet implemented. Not yet implemented.

Arithmetic

Count 11/12

NNabla Function Status Description

Add2 BcAdd2 Sub2 Mul2 Div2 Pow2 AddScalar MulScalar PowScalar RSubScalar RDivScalar RPowScalar

✓

✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓

Not yet implemented.

Logical

Count 29/29

NNabla Function Status Description

Sign Minimum2 Maximum2 MinimumScalar MaximumScalar LogicalAnd LogicalOr LogicalXor Equal NotEqual GreaterEqual Greater LessEqual Less LogicalAndScalar LogicalOrScalar LogicalXorScalar EqualScalar NotEqualScalar GreaterEqualScalar GreaterScalar LessEqualScalar LessScalar LogicalNot IsNaN IsInf ResetNaN ResetInf Where

✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓

Math

Count 22/22

NNabla Function Status Description

Constant Arange Abs Exp Log Identity BatchMatmul Round Ceil Floor Sin Cos Tan Sinh Cosh ASin ACos ATan ATan2 ASinh ACosh ATanh

✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓

Array Manipulation

Count 12/19

NNabla Function Status Description

Concatenate Split Stack Slice

✓ ✓ ✓ ✓

Pad Transpose Broadcast BroadcastTo Tile OneHot Flip Shift Sort Reshape MatrixDiag MatrixDiagPart Assign GatherNd ScatterNd

△ ✓ ✓ ✓ ✓ ✓ ✓

✓

When the mode of the pad is reflect, if the size of the pad exceeds the input size, tensorflow cannot handle it.

Not yet implemented. Not yet implemented.

Not yet implemented. Not yet implemented. Not yet implemented. Not yet implemented. Not yet implemented.

Signal Processing

Count 1/3

NNabla Function Status Description

Interpolate FFT IFFT

✓

Not yet implemented. Not yet implemented.

Stochasticity

Count 0/11

NNabla Function Status Description

Dropout TopKData TopKGrad Rand Randint Randn RandomChoice RandomCrop RandomFlip RandomShift ImageAugmentation

X

The Dropout in nnabla has no test mode and contains random parameters, so the test result is not the same as tensorflow. Not yet implemented. Not yet implemented. Not yet implemented. Not yet implemented. Not yet implemented. Not yet implemented. Not yet implemented. Not yet implemented. Not yet implemented. Not yet implemented.

Loss Functions

Count 0/9

NNabla Function Status Description

SigmoidCrossEntropy BinaryCrossEntropy SoftmaxCrossEntropy CategoricalCrossEntropy SquaredError AbsoluteError HuberLoss EpsilonInsensitiveLoss KLMultinomial

Not yet implemented. Not yet implemented. Not yet implemented. Not yet implemented. Not yet implemented. Not yet implemented. Not yet implemented. Not yet implemented. Not yet implemented.

Quantization Neural Network Layers

Count 6/12

NNabla Function Status Description

BinarySigmoid BinaryTanh BinaryConnectAffine

✓ ✓ ✓

BinaryConnectConvolution BinaryWeightAffine

△ ✓

The cases dilations and strides larger than 1 are not supported by tensorflow.

BinaryWeightConvolution INQAffine INQConvolution FixedPointQuantize MinMaxQuantize Pow2Quantize Prune

△

The cases dilations and strides larger than 1 are not supported by tensorflow. Not yet implemented. Not yet implemented. Not yet implemented. Not yet implemented. Not yet implemented. Not yet implemented.

Validation

Count 0/3

NNabla Function Status Description

TopNError BinaryError ConfusionMatrix

Not yet implemented. Not yet implemented. Not yet implemented.

Unsupported, Special Use

Count 0/5

NNabla Function Status Description

VATNoise Unlink Sink NmsDetection2d MaxPoolingBackward

Not yet implemented. Not yet implemented. Not yet implemented. Not yet implemented. Not yet implemented.

Tensorflow Lite Support Status

Export

✓: Supported
△: Partially supported
X: Supported, but test failed.
Empty: Not support yet.

Total: 98/173

Neural Network Layer

Count 8/14

NNabla Function Status

Affine RNN LSTM GRU

✓

Convolution

△

DepthwiseConvolution

△

Deconvolution

△

DepthwiseDeconvolution

△

MaxPooling

X

AveragePooling

X

GlobalAveragePooling

✓

SumPooling

X

Unpooling

△

Embed

✓

Neural Network Activation Functions

Count 20/21

NNabla Function Status

Sigmoid

✓

Swish

✓

Tanh

✓

ReLU

✓

LeakyReLU

✓

Softmax

✓

LogSoftmax

✓

ELU

✓

SELU

△

CReLU

✓

CELU

✓

PReLU

✓

GELU

✓

ReLU6

✓

HardSigmoid

✓

HardTanh

✓

LogSigmoid

✓

SoftPlus

✓

SoftSign

✓

TanhShrink

✓

Sinc

X

Normalization

Count 0/6

NNabla Function Status

FusedBatchNormalization

X

BatchNormalization SyncBatchNormalization MeanSubtraction ClipGradByValue ClipGradByNorm

X

Reduction

Count 5/7

NNabla Function Status

Sum

✓

Mean

✓

Max

✓

Min

✓

Prod ReduceSum ReduceMean

✓

Arithmetic

Count 11/12

NNabla Function Status

Add2 BcAdd2

✓

Sub2

✓

Mul2

✓

Div2

✓

Pow2

✓

AddScalar

✓

MulScalar

✓

PowScalar

✓

RSubScalar

✓

RDivScalar

✓

RPowScalar

✓

Logical

Count 25/29

NNabla Function Status

Sign

✓

Minimum2

✓

Maximum2

✓

MinimumScalar

✓

MaximumScalar

✓

LogicalAnd

✓

LogicalOr

✓

LogicalXor

✓

Equal

✓

NotEqual

✓

GreaterEqual

✓

Greater

✓

LessEqual

✓

Less

✓

LogicalAndScalar

✓

LogicalOrScalar

✓

LogicalXorScalar

✓

EqualScalar

✓

NotEqualScalar

✓

GreaterEqualScalar

✓

GreaterScalar

✓

LessEqualScalar

✓

LessScalar

✓

LogicalNot

✓

IsNaN

✓

IsInf

X

ResetNaN

X

ResetInf

X

Where

X

Math

Count 14/22

NNabla Function Status

Constant

✓

Arange

✓

Abs

✓

Exp

✓

Log

✓

Identity

✓

BatchMatmul

✓

Round

X

Ceil

✓

Floor

✓

Sin

✓

Cos

✓

Tan

✓

Sinh

✓

Cosh

✓

ASin

X

ACos

X

ATan

X

ATan2

X

ASinh

X

ACosh

X

ATanh

X

Array Manipulation

Count 11/19

NNabla Function Status

Concatenate

✓

Split

✓

Stack

✓

Slice

△

Pad

X

Transpose

✓

Broadcast

✓

BroadcastTo

✓

Tile

✓

OneHot

✓

Flip Shift Sort

✓

Reshape MatrixDiag MatrixDiagPart Assign GatherNd ScatterNd

✓

Signal Processing

Count 0/3

NNabla Function Status

Interpolate FFT IFFT

X

Stochasticity

Count 0/11

NNabla Function Status

Dropout TopKData TopKGrad Rand Randint Randn RandomChoice RandomCrop RandomFlip RandomShift ImageAugmentation

X

Loss Functions

Count 0/9

NNabla Function Status

SigmoidCrossEntropy BinaryCrossEntropy SoftmaxCrossEntropy CategoricalCrossEntropy SquaredError AbsoluteError HuberLoss EpsilonInsensitiveLoss KLMultinomial

Quantization Neural Network Layers

Count 4/12

NNabla Function Status

BinarySigmoid

X

BinaryTanh

X

BinaryConnectAffine

✓

BinaryConnectConvolution

△

BinaryWeightAffine

✓

BinaryWeightConvolution INQAffine INQConvolution FixedPointQuantize MinMaxQuantize Pow2Quantize Prune

△

Validation

Count 0/3

NNabla Function Status

TopNError BinaryError ConfusionMatrix

Unsupported, Special Use

Count 0/5

NNabla Function Status

VATNoise Unlink Sink NmsDetection2d MaxPoolingBackward

NNabla C Runtime Support Status

NNabla version: None

✓: Supported
△: Partially supported
X: Supported, but test failed or no test data.
Empty: Not support yet.

Export

Total: 56/173

Neural Network Layer

Count 8/14

NNabla Function Status Description

Affine RNN LSTM GRU Convolution DepthwiseConvolution Deconvolution DepthwiseDeconvolution MaxPooling AveragePooling GlobalAveragePooling SumPooling Unpooling Embed

✓

✓ ✓ ✓

✓ ✓

✓ ✓

Neural Network Activation Functions

Count 11/21

NNabla Function Status Description

Sigmoid Swish Tanh ReLU LeakyReLU Softmax LogSoftmax ELU SELU CReLU CELU PReLU GELU ReLU6 HardSigmoid HardTanh LogSigmoid SoftPlus SoftSign TanhShrink Sinc

✓ ✓ ✓ ✓ ✓ ✓

✓ ✓ ✓ ✓ ✓

Normalization

Count 1/6

NNabla Function Status Description

FusedBatchNormalization BatchNormalization SyncBatchNormalization MeanSubtraction ClipGradByValue ClipGradByNorm

✓

X

Reduction

Count 1/7

NNabla Function Status Description

Sum Mean Max Min Prod ReduceSum ReduceMean

✓

Arithmetic

Count 11/12

NNabla Function Status Description

Add2 BcAdd2 Sub2 Mul2 Div2 Pow2 AddScalar MulScalar PowScalar RSubScalar RDivScalar RPowScalar

✓

✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓

Logical

Count 5/29

NNabla Function Status Description

Sign Minimum2 Maximum2 MinimumScalar MaximumScalar LogicalAnd LogicalOr LogicalXor Equal NotEqual GreaterEqual Greater LessEqual Less LogicalAndScalar LogicalOrScalar LogicalXorScalar EqualScalar NotEqualScalar GreaterEqualScalar GreaterScalar LessEqualScalar LessScalar LogicalNot IsNaN IsInf ResetNaN ResetInf Where

✓ ✓ ✓ ✓ ✓

Math

Count 6/22

NNabla Function Status Description

Constant Arange Abs Exp Log Identity BatchMatmul Round Ceil Floor Sin Cos Tan Sinh Cosh ASin ACos ATan ATan2 ASinh ACosh ATanh

✓ ✓ ✓ ✓ △ ✓

Array Manipulation

Count 7/19

NNabla Function Status Description

Concatenate Split Stack Slice Pad Transpose Broadcast BroadcastTo Tile OneHot Flip Shift Sort Reshape MatrixDiag MatrixDiagPart Assign GatherNd ScatterNd

✓ ✓ ✓ ✓

✓

✓ X

✓ X X

Signal Processing

Count 0/3

NNabla Function Status Description

Interpolate FFT IFFT

Stochasticity

Count 0/11

NNabla Function Status Description

Dropout TopKData TopKGrad Rand Randint Randn RandomChoice RandomCrop RandomFlip RandomShift ImageAugmentation

X

Loss Functions

Count 0/9

NNabla Function Status Description

SigmoidCrossEntropy BinaryCrossEntropy SoftmaxCrossEntropy CategoricalCrossEntropy SquaredError AbsoluteError HuberLoss EpsilonInsensitiveLoss KLMultinomial

Quantization Neural Network Layers

Count 6/12

NNabla Function Status Description

BinarySigmoid BinaryTanh BinaryConnectAffine BinaryConnectConvolution BinaryWeightAffine BinaryWeightConvolution INQAffine INQConvolution FixedPointQuantize MinMaxQuantize Pow2Quantize Prune

✓ ✓ ✓ ✓ ✓ ✓

Validation

Count 0/3

NNabla Function Status Description

TopNError BinaryError ConfusionMatrix

Unsupported, Special Use

Count 0/5

NNabla Function Status Description

VATNoise Unlink Sink NmsDetection2d MaxPoolingBackward

NNabla Function	Status	Description
Affine RNN LSTM GRU	✓	Not yet implemented. Not yet implemented. Not yet implemented.
Convolution	△	The cases dilations and strides larger than 1 are not supported by tensorflow.
DepthwiseConvolution	△	The cases dilations and strides larger than 1 are not supported by tensorflow.
Deconvolution	△	The cases dilations larger than 1 are not supported by tensorflow.
DepthwiseDeconvolution MaxPooling	△ ✓	The cases dilations larger than 1 are not supported by tensorflow.
AveragePooling GlobalAveragePooling SumPooling	△ ✓ ✓	Currently only supports the cases both ignore_border and including_pad are True.
Unpooling Embed	△ ✓	The kernel only supports 2d.

NNabla Function	Status	Description
BinarySigmoid BinaryTanh BinaryConnectAffine	✓ ✓ ✓
BinaryConnectConvolution BinaryWeightAffine	△ ✓	The cases dilations and strides larger than 1 are not supported by tensorflow.
BinaryWeightConvolution INQAffine INQConvolution FixedPointQuantize MinMaxQuantize Pow2Quantize Prune	△	The cases dilations and strides larger than 1 are not supported by tensorflow. Not yet implemented. Not yet implemented. Not yet implemented. Not yet implemented. Not yet implemented. Not yet implemented.

NNabla Function	Status
Sigmoid	✓
Swish	✓
Tanh	✓
ReLU	✓
LeakyReLU	✓
Softmax	✓
LogSoftmax	✓
ELU	✓
SELU	△
CReLU	✓
CELU	✓
PReLU	✓
GELU	✓
ReLU6	✓
HardSigmoid	✓
HardTanh	✓
LogSigmoid	✓
SoftPlus	✓
SoftSign	✓
TanhShrink	✓
Sinc	X

NNabla Function	Status
FusedBatchNormalization	X
BatchNormalization SyncBatchNormalization MeanSubtraction ClipGradByValue ClipGradByNorm	X

NNabla Function	Status
Add2 BcAdd2	✓
Sub2	✓
Mul2	✓
Div2	✓
Pow2	✓
AddScalar	✓
MulScalar	✓
PowScalar	✓
RSubScalar	✓
RDivScalar	✓
RPowScalar	✓

NNabla Function	Status
Sign	✓
Minimum2	✓
Maximum2	✓
MinimumScalar	✓
MaximumScalar	✓
LogicalAnd	✓
LogicalOr	✓
LogicalXor	✓
Equal	✓
NotEqual	✓
GreaterEqual	✓
Greater	✓
LessEqual	✓
Less	✓
LogicalAndScalar	✓
LogicalOrScalar	✓
LogicalXorScalar	✓
EqualScalar	✓
NotEqualScalar	✓
GreaterEqualScalar	✓
GreaterScalar	✓
LessEqualScalar	✓
LessScalar	✓
LogicalNot	✓
IsNaN	✓
IsInf	X
ResetNaN	X
ResetInf	X
Where	X

NNabla Function	Status
Constant	✓
Arange	✓
Abs	✓
Exp	✓
Log	✓
Identity	✓
BatchMatmul	✓
Round	X
Ceil	✓
Floor	✓
Sin	✓
Cos	✓
Tan	✓
Sinh	✓
Cosh	✓
ASin	X
ACos	X
ATan	X
ATan2	X
ASinh	X
ACosh	X
ATanh	X

NNabla Function	Status
Concatenate	✓
Split	✓
Stack	✓
Slice	△
Pad	X
Transpose	✓
Broadcast	✓
BroadcastTo	✓
Tile	✓
OneHot	✓
Flip Shift Sort	✓
Reshape MatrixDiag MatrixDiagPart Assign GatherNd ScatterNd	✓

NNabla Function	Status
BinarySigmoid	X
BinaryTanh	X
BinaryConnectAffine	✓
BinaryConnectConvolution	△
BinaryWeightAffine	✓
BinaryWeightConvolution INQAffine INQConvolution FixedPointQuantize MinMaxQuantize Pow2Quantize Prune	△

Files

Function-Level_Support_Status.rst

Latest commit

History

Function-Level_Support_Status.rst

File metadata and controls

Function-Level Support Status

ONNX Support Status

Import

Export

Neural Network Layer

Neural Network Activation Functions

Normalization

Reduction

Arithmetic

Logical

Math

Array Manipulation

Signal Processing

Stochasticity

Loss Functions

Quantization Neural Network Layers

Validation

Unsupported, Special Use

Tensorflow Support Status

Import

Export

Neural Network Layer

Neural Network Activation Functions

Normalization

Reduction

Arithmetic

Logical

Math

Array Manipulation

Signal Processing

Stochasticity

Loss Functions

Quantization Neural Network Layers

Validation

Unsupported, Special Use

Tensorflow Lite Support Status

Export

Neural Network Layer

Neural Network Activation Functions

Normalization

Reduction

Arithmetic

Logical

Math

Array Manipulation

Signal Processing

Stochasticity

Loss Functions

Quantization Neural Network Layers

Validation

Unsupported, Special Use

NNabla C Runtime Support Status

Export

Neural Network Layer

Neural Network Activation Functions

Normalization

Reduction

Arithmetic

Logical

Math

Array Manipulation

Signal Processing

Stochasticity

Loss Functions

Quantization Neural Network Layers

Validation

Unsupported, Special Use