Minimal tested opset version 9, maximum tested opset version 16, recommended opset version 13
| Operation type | Supported | Restrictions |
|---|---|---|
| Abs | Y | |
| Acos | Y | |
| Acosh | N | |
| Add | Y | |
| And | Y | |
| ArgMax | N | |
| ArgMin | N | |
| Asin | Y | |
| Asinh | N | |
| Atan | Y | |
| Atanh | N | |
| AveragePool | Y | Average pool operation with spatial rank > 3 is not implemented |
| BatchNormalization | Y | BatchNorm operation with spatial rank > 3 is not implemented. BatchNorm nodes in training mode are not supported |
| BitShift | N | |
| Cast | Y | |
| Ceil | Y | |
| Clip | Y | Dynamic value of min/max is not implemented |
| Compress | N | |
| Concat | Y | |
| ConcatFromSequence | N | |
| Constant | Y | |
| ConstantOfShape | Y | Parameter "value" must be scalar |
| Conv | Y | Convolution operation with spatial rank > 3 is not implemented |
| ConvInteger | N | |
| ConvTranspose | Y | Convolution operation with spatial rank > 3 is not implemented |
| Cos | Y | |
| Cosh | N | |
| CumSum | Y | |
| DepthToSpace | Y | DCR mode is not implemented |
| DequantizeLinear | N | |
| Det | N | |
| Div | Y | |
| Dropout | Y | |
| Einsum | Y | |
| Elu | Y | |
| Equal | Y | |
| Erf | Y | |
| Exp | Y | |
| Expand | Y | |
| EyeLike | Y | |
| Flatten | Y | |
| Floor | Y | |
| GRU | N | |
| Gather | Y | |
| GatherElements | Y | |
| GatherND | Y | GatherND operation with parameter "batch_dims" > 0 is not implemented |
| Gemm | Y | |
| GlobalAveragePool | Y | |
| GlobalLpPool | N | |
| GlobalMaxPool | N | |
| Greater | Y | |
| GridSample | N | |
| HardSigmoid | Y | |
| Hardmax | N | |
| Identity | Y | |
| If | N | |
| InstanceNormalization | Y | |
| IsInf | Y | |
| IsNaN | Y | |
| LayerNormalization | Y | LayerNormalization outputs "Mean" and "InvStdDev" are not implemented |
| LRN | Y | |
| LSTM | N | |
| LeakyRelu | Y | |
| Less | Y | |
| Log | Y | |
| Loop | N | |
| LpNormalization | N | |
| LpPool | N | |
| MatMul | Y | |
| MatMulInteger | N | |
| Max | Y | |
| MaxPool | Y | Max pool operation with spatial rank > 3 is not implemented |
| MaxRoiPool | N | |
| MaxUnpool | N | |
| Mean | Y | |
| Min | Y | |
| Mod | Y | |
| Mul | Y | |
| Multinomial | N | |
| Neg | Y | |
| NonMaxSuppression | Y | |
| NonZero | Y | |
| Not | Y | |
| OneHot | N | |
| Optional | N | |
| OptionalGetElement | N | |
| OptionalHasElement | N | |
| Or | Y | |
| PRelu | Y | |
| Pad | Y | Padding is implemented to pad the last 3 dimensions of 5D input tensor, or the last 2 dimensions of 4D input tensor, or the last dimension of 3D input tensor |
| Pow | Y | |
| QLinearConv | N | |
| QLinearMatMul | N | |
| QuantizeLinear | N | |
| RNN | N | |
| RandomNormal | N | |
| RandomNormalLike | N | |
| RandomUniform | N | |
| RandomUniformLike | N | |
| Reciprocal | Y | |
| ReduceL1 | Y | |
| ReduceL2 | Y | |
| ReduceLogSum | Y | |
| ReduceLogSumExp | Y | |
| ReduceMax | Y | |
| ReduceMean | Y | |
| ReduceMin | Y | |
| ReduceProd | Y | |
| ReduceSum | Y | |
| ReduceSumSquare | Y | |
| Relu | Y | |
| Reshape | Y | Parameter "allowzero" = 1 is not implemented |
| Resize | Y | Roi logic is not implemented (pytorch's interpolate cannot resize channel or batch dimensions) |
| ReverseSequence | N | |
| RoiAlign | Y | Only "avg" mode is supported |
| Round | Y | |
| Scan | N | |
| Scatter(deprecated) | N | |
| ScatterElements | N | |
| ScatterND | Y | Only "none" reduction is supported |
| Selu | Y | Parameters "alpha" and "gamma" must be default |
| SequenceAt | N | |
| SequenceConstruct | N | |
| SequenceEmpty | N | |
| SequenceErase | N | |
| SequenceInsert | N | |
| SequenceLength | N | |
| Shape | Y | |
| Shrink | N | |
| Sigmoid | Y | |
| Sign | Y | |
| Sin | Y | |
| Sinh | N | |
| Size | N | |
| Slice | Y | |
| Softplus | Y | |
| Softsign | Y | |
| SpaceToDepth | N | |
| Split | Y | |
| SplitToSequence | N | |
| Sqrt | Y | |
| Squeeze | Y | |
| StringNormalizer | N | |
| Sub | Y | |
| Sum | Y | |
| Tan | Y | |
| Tanh | Y | |
| TfIdfVectorizer | N | |
| ThresholdedRelu | N | |
| Tile | Y | |
| TopK | Y | |
| Transpose | Y | |
| Trilu | N | |
| Unique | N | |
| Unsqueeze | Y | |
| Upsample(deprecated) | N | |
| Where | Y | |
| Xor | Y | |
| Function | N | |
| Bernoulli | N | |
| CastLike | N | |
| Celu | Y | |
| DynamicQuantizeLinear | N | |
| GreaterOrEqual | Y | |
| HardSwish | Y | |
| LessOrEqual | Y | |
| LogSoftmax | Y | |
| MeanVarianceNormalization | N | |
| NegativeLogLikelihoodLoss | N | |
| Range | Y | |
| SequenceMap | N | |
| Softmax | Y |