Supported op lowerings

coreai-onnx currently supports 143 built-in op lowerings. The table below lists every ONNX, Core AI composite, and internal fused op key that can be converted without any extra code.

If your model uses ops not listed here, you can add support via a custom lowering — see Custom lowerings for a worked example.

Op key	Status
Abs	supported
Acos	supported
Acosh	supported
Add	supported
And	supported
ArgMax	supported
ArgMin	supported
Asin	supported
Asinh	supported
Atan	supported
Atanh	supported
AveragePool	supported
BatchNormalization	supported
BitwiseAnd	supported
BitwiseNot	supported
BitwiseOr	supported
BitwiseXor	supported
Cast	supported
CastLike	supported
Ceil	supported
Celu	supported
Clip	supported
Compress	supported
Concat	supported
Constant	supported
ConstantOfShape	supported
Conv	supported
ConvTranspose	supported
Cos	supported
Cosh	supported
CumSum	supported
DepthToSpace	supported
DequantizeLinear	supported
Div	supported
Dropout	supported
Einsum	supported
Elu	supported
Equal	supported
Erf	supported
Exp	supported
Expand	supported
EyeLike	supported
Flatten	supported
Floor	supported
Gather	supported
GatherElements	supported
GatherND	supported
Gelu	supported
Gemm	supported
GlobalAveragePool	supported
GlobalLpPool	supported
GlobalMaxPool	supported
Greater	supported
GreaterOrEqual	supported
GridSample	supported
GroupNormalization	supported
HardSigmoid	supported
HardSwish	supported
Hardmax	supported
Identity	supported
If	supported
InstanceNormalization	supported
IsInf	supported
IsNaN	supported
LRN	supported
LSTM	supported
LayerNormalization	supported
LeakyRelu	supported
Less	supported
LessOrEqual	supported
Log	supported
LogSoftmax	supported
LpNormalization	supported
LpPool	supported
MatMul	supported
Max	supported
MaxPool	supported
Mean	supported
MeanVarianceNormalization	supported
Min	supported
Mish	supported
Mod	supported
Mul	supported
Neg	supported
NonZero	supported
Not	supported
OneHot	supported
Or	supported
PRelu	supported
Pad	supported
Pow	supported
QLinearConv	supported
QLinearMatMul	supported
QuantizeLinear	supported
Range	supported
Reciprocal	supported
ReduceL1	supported
ReduceL2	supported
ReduceLogSum	supported
ReduceLogSumExp	supported
ReduceMax	supported
ReduceMean	supported
ReduceMin	supported
ReduceProd	supported
ReduceSum	supported
ReduceSumSquare	supported
Relu	supported
Reshape	supported
Resize	supported
Round	supported
ScatterElements	supported
ScatterND	supported
Selu	supported
Shape	supported
Shrink	supported
Sigmoid	supported
Sign	supported
Sin	supported
Sinh	supported
Size	supported
Slice	supported
Softmax	supported
Softplus	supported
Softsign	supported
SpaceToDepth	supported
Split	supported
Sqrt	supported
Squeeze	supported
Sub	supported
Sum	supported
Tan	supported
Tanh	supported
ThresholdedRelu	supported
Tile	supported
TopK	supported
Transpose	supported
Trilu	supported
Unsqueeze	supported
Where	supported
Xor	supported
com.microsoft::Inverse	supported
coreai::ScaledDotProductAttention	supported
coreai_onnx::Silu	supported