Supported ONNX operators

The table shows which Open Neural Network Exchange (ONNX) operators Sentis supports, and which data types Sentis supports for each back end type.

When you import a model, each ONNX operator in the model graph becomes a Sentis layer. A Sentis layer has the same name as the ONNX operator, unless the table shows the operator maps to a different layer. Refer to How Sentis optimizes a model for more information.

If you use a GPU Worker and the model has a layer that isn't supported, or has a layer that uses an input tensor data type that isn't supported, Sentis falls back to using BackendType.CPU for the layer, which uses Burst. If BackendType.CPU isn't supported, Sentis falls back to using a slower CPU implementation of the layer that doesn't use Burst.

Refer to the API reference for more information on each Sentis layer.

Supported operators

ONNX operators

Name Supported data types with BackendType.CPU Supported data types with BackendType.GPUCompute Supported data types with BackendType.GPUPixel Notes
Abs float, int float, int float
Acos float float float
Acosh float float float
Add float, int float, int float
And int int Not supported
ArgMax float, int float, int Not supported
ArgMin float, int float, int Not supported
Asin float float float
Asinh float float float
Atan float float float
Atanh float float float
AveragePool float float (1D and 2D only) float (2D only) The ceil_mode and count_include_pad parameters aren't supported.
BatchNormalization float float float The epsilon, momentum, spatial and training_mode parameters aren't supported.
Bernoulli float float Not supported
Cast float, int float, int Not supported
CastLike float, int float, int Not supported
Ceil float float float
Celu float float float
Clip float float float
Compress float, int Not supported Not supported
Concat float, int float, int float
Constant - - - The sparse_value parameter isn't supported.
ConstantOfShape float, int float, int float
Conv float float (1D, 2D and 3D) float (2D only) The kernel_shape parameter isn't supported.
ConvTranspose float float (2D only) float (2D only) The dilations, group, kernel_shape and output_shape parameters aren't supported.
Cos float float float
Cosh float float float
CumSum float, int float, int Not supported
DepthToSpace float float float
Div float, int float, int float
Dropout - - - The operator maps to the Sentis layer Identity
Einsum float float (1 or 2 inputs only) Not supported
Elu float float float
Equal float, int float, int Not supported
Erf float float float
Exp float float float
Expand float, int float, int float
Flatten float, int float, int float
Floor float float float
Gather float, int float, int Not supported
GatherElements float, int float, int Not supported
GatherND float, int float, int Not supported
Gemm float float float
GlobalAveragePool float float float
GlobalMaxPool float float float
Greater float, int float, int Not supported
GreaterOrEqual float, int float, int Not supported
Hardmax float float Not supported
HardSigmoid float float float
HardSwish float float float
Identity float, int float, int float
InstanceNormalization float float Not supported
IsInf float float Not supported
IsNaN float float Not supported
LeakyRelu float float float
Less float, int float, int Not supported
LessOrEqual float, int float, int Not supported
Log float float float
LogSoftmax float float float
LRN float Not supported Not supported
LSTM float float Not supported
MatMul float float float (2D only)
Max float, int float, int float
MaxPool float float (1D and 2D only) float (2D only) The ceil_mode, dilations and storage_order parameters aren't supported.
Mean float float float
Min float, int float, int float
Mod float, int float, int float
Mul float, int float, int float
Multinomial float Not supported Not supported
Neg float, int float, int float
NonMaxSuppression float Not supported Not supported
NonZero float, int Not supported Not supported
Not int int Not supported
OneHot (int, int, int) - outputs int (int, int, int) - outputs int Not supported
Or int int Not supported
Pad float float float
Pow (float float) or (float, int) (float float) or (float, int) float
PRelu float float Not supported
RandomNormal float float Not supported
RandomNormalLike float float Not supported
RandomUniform float float Not supported
RandomUniformLike float float Not supported
Range float float Not supported
Reciprocal float float float
ReduceL1 float, int float, int float
ReduceL2 float float float
ReduceLogSum float float float
ReduceLogSumExp float float float
ReduceMax float, int float, int float
ReduceMean float float float
ReduceMin float, int float, int float
ReduceProd float, int float, int float
ReduceSum float, int float, int float
ReduceSumSquare float, int float, int float
Relu float float float
Reshape float, int float, int float
Resize float float (2D and 3D only) float (2D only) The cubic_coeff_a, exclude_outside, extrapolation_value and roi parameters aren't supported.
RoiAlign float float Not supported The coordinate_transformation_mode and mode parameters aren't supported.
Round float float float
Scatter (deprecated) float, int float, int Not supported The operator maps to the Sentis layer ScatterElements.
ScatterElements float, int float, int (no ScatterReductionMode) Not supported
ScatterND float float Not supported
Selu float float float
Shape - - -
Shrink float float Not supported
Sigmoid float float float
Sign float, int float, int Not supported
Sin float float float
Sinh float float float
Size - - -
Slice float, int float, int float
Softmax float float float
Softplus float float float
Softsign float float float
SpaceToDepth float float float
Split float, int float, int float
Sqrt float float float
Squeeze float, int float, int float
Sub float, int float, int float
Sum float, int float, int Not supported
Tan float float float
Tanh float float float
ThresholdedRelu float float float
Tile float, int float, int Not supported
TopK float Not supported Not supported
Transpose float, int float, int float
Trilu float, int float, int Not supported
Unsqueeze float, int float, int float
Upsample (deprecated) float float (2D and 3D only) float (2D only) The operator maps to the Sentis layer Resize.
Where (int, float, float) - outputs float, (int, int, int) - outputs int (int, float, float) - outputs float, (int, int, int) - outputs int Not supported
Xor int int Not supported

Sentis-only layers

Sentis might create the following layers when it optimizes the model.

Name Supported data types with BackendType.CPU Supported data types with BackendType.GPUCompute Supported data types with BackendType.GPUPixel
Dense float float Not supported
MatMul2D float float float
Relu6 float float float
Square float float float
Swish float float float

Unsupported operators

The following ONNX operators aren't supported in the current version of Sentis.

  • BitShift
  • ConcatFromSequence
  • ConvInteger
  • DequantizeLinear
  • Det
  • DynamicQuantizeLinear
  • EyeLike
  • If
  • GRU
  • Loop
  • LpPool
  • MatMulInteger
  • MaxUnpool
  • MeanVarianceNormalization
  • NegativeLogLikelihoodLoss
  • Optional
  • OptionalGetElement
  • OptionalHasElement
  • QLinearConv
  • QLinearMatMul
  • QuantizeLinear
  • ReverseSequence
  • RNN
  • Scan
  • SequenceAt
  • SequenceConstruct
  • SequenceEmpty
  • SequenceErase
  • SequenceInsert
  • SequenceLength
  • SoftmaxCrossEntropyLoss
  • SplitToSequence
  • StringNormalizer
  • TfIdfVectorizer
  • Unique

Additional resources