Comparison: onnx OpSchema vs onnx_light LightOpSchema

This page programmatically compares the operator schemas exposed by the reference onnx package (onnx.defs.OpSchema) with the lightweight schemas exposed by onnx_light (LightOpSchema, from the C++ onnx_op extension).

For every operator known to either side the table below reports:

• whether the operator is described by an OpSchema (onnx);

• whether the operator is described by a LightOpSchema (onnx_light);

• whether a shape-inference function is registered on the onnx side (OpSchema.has_type_and_shape_inference_function);

• whether a shape-inference function is registered on the onnx_light side. Shape inference for onnx_light lives in the onnx_optim library (onnx_light/onnx_optim/shapes/shape_inference.cc), separate from the schema definitions themselves;

• how many node backend tests exercise the operator in each package. A test case is attributed to the op_type of the first node of its model — the convention used by the ONNX node backend test suite, and also the one used by the test cases generated by onnx_light.backend.test.case.base.collect_test_case().

Note

The page is regenerated every time the documentation is built, so the numbers always reflect the operator schemas and backend tests shipped with the installed onnx and onnx_light packages. When the reference onnx package is not installed, the onnx-side columns are reported as empty / zero.

---

Summary

Metric	onnx	onnx_light
Operators with a schema	223	73
Operators with shape inference	212	23
Node backend tests (counted)	1678	105

---

Per-operator comparison

The table below is rendered with the sphinx-datatables extension, so it is interactive: use the search box to filter operators by any column (domain, name, yes/no flags, test counts) and click on a column header to sort the table by that column (click again to toggle ascending / descending order).

Domain	Operator	onnx	onnx_light	onnx shape inference	onnx_light shape inference (onnx_optim)	onnx backend tests	onnx_light backend tests
ai.onnx	Abs	yes	yes	yes	yes	10	1
ai.onnx	Acos	yes	yes	yes	yes	2	1
ai.onnx	Acosh	yes	yes	yes	yes	2	1
ai.onnx	Add	yes	yes	yes	yes	8	2
ai.onnx	AffineGrid	yes	no	yes	no	4	0
ai.onnx	And	yes	yes	yes	yes	8	10
ai.onnx	ArgMax	yes	yes	yes	no	16	0
ai.onnx	ArgMin	yes	yes	yes	no	16	0
ai.onnx	Asin	yes	yes	yes	no	2	0
ai.onnx	Asinh	yes	yes	yes	no	2	0
ai.onnx	Atan	yes	no	yes	no	2	0
ai.onnx	Atanh	yes	no	yes	no	2	0
ai.onnx	Attention	yes	no	yes	no	62	0
ai.onnx	AveragePool	yes	yes	yes	yes	20	11
ai.onnx	BatchNormalization	yes	no	yes	no	4	0
ai.onnx	Bernoulli	yes	no	yes	no	3	0
ai.onnx	BitCast	yes	no	yes	no	10	0
ai.onnx	BitShift	yes	no	yes	no	8	0
ai.onnx	BitwiseAnd	yes	no	yes	no	4	0
ai.onnx	BitwiseNot	yes	no	yes	no	3	0
ai.onnx	BitwiseOr	yes	no	yes	no	4	0
ai.onnx	BitwiseXor	yes	no	yes	no	4	0
ai.onnx	BlackmanWindow	yes	yes	yes	no	2	2
ai.onnx	Cast	yes	yes	yes	yes	125	4
ai.onnx	CastLike	yes	no	yes	no	56	0
ai.onnx	Ceil	yes	no	yes	no	2	0
ai.onnx	Celu	yes	no	yes	no	1	0
ai.onnx	CenterCropPad	yes	no	yes	no	6	0
ai.onnx	Clip	yes	no	yes	no	12	0
ai.onnx	Col2Im	yes	no	yes	no	5	0
ai.onnx	Compress	yes	no	yes	no	4	0
ai.onnx	Concat	yes	yes	yes	yes	12	2
ai.onnx	ConcatFromSequence	yes	yes	yes	no	0	0
ai.onnx	Constant	yes	yes	yes	yes	197	1
ai.onnx	ConstantOfShape	yes	no	yes	no	3	0
ai.onnx	Conv	yes	no	yes	no	6	0
ai.onnx	ConvInteger	yes	no	yes	no	2	0
ai.onnx	ConvTranspose	yes	no	yes	no	11	0
ai.onnx	Cos	yes	yes	yes	no	2	0
ai.onnx	Cosh	yes	yes	yes	no	2	0
ai.onnx	CumProd	yes	no	yes	no	9	0
ai.onnx	CumSum	yes	no	yes	no	9	0
ai.onnx	DFT	yes	no	yes	no	10	0
ai.onnx	DeformConv	yes	no	yes	no	4	0
ai.onnx	DepthToSpace	yes	no	yes	no	2	0
ai.onnx	DequantizeLinear	yes	no	yes	no	14	0
ai.onnx	Det	yes	no	yes	no	2	0
ai.onnx	Div	yes	yes	yes	no	10	5
ai.onnx	Dropout	yes	no	yes	no	12	0
ai.onnx	DynamicQuantizeLinear	yes	no	yes	no	3	0
ai.onnx	Einsum	yes	no	yes	no	6	0
ai.onnx	Elu	yes	no	yes	no	3	0
ai.onnx	Equal	yes	yes	yes	no	10	0
ai.onnx	Erf	yes	no	yes	no	1	0
ai.onnx	Exp	yes	no	yes	no	4	0
ai.onnx	Expand	yes	no	yes	no	2	0
ai.onnx	EyeLike	yes	no	yes	no	3	0
ai.onnx	Flatten	yes	no	yes	no	9	0
ai.onnx	Floor	yes	no	yes	no	2	0
ai.onnx	GRU	yes	yes	yes	no	4	0
ai.onnx	Gather	yes	no	yes	no	4	0
ai.onnx	GatherElements	yes	no	yes	no	3	0
ai.onnx	GatherND	yes	no	yes	no	3	0
ai.onnx	Gelu	yes	no	yes	no	4	0
ai.onnx	Gemm	yes	yes	yes	no	11	0
ai.onnx	GlobalAveragePool	yes	no	yes	no	2	0
ai.onnx	GlobalLpPool	yes	no	yes	no	0	0
ai.onnx	GlobalMaxPool	yes	no	yes	no	2	0
ai.onnx	Greater	yes	yes	yes	yes	16	4
ai.onnx	GreaterOrEqual	yes	no	yes	no	8	0
ai.onnx	GridSample	yes	no	yes	no	18	0
ai.onnx	GroupNormalization	yes	no	no	no	2	0
ai.onnx	HammingWindow	yes	no	yes	no	2	0
ai.onnx	HannWindow	yes	no	yes	no	2	0
ai.onnx	HardSigmoid	yes	no	yes	no	4	0
ai.onnx	HardSwish	yes	no	yes	no	1	0
ai.onnx	Hardmax	yes	no	yes	no	7	0
ai.onnx	Identity	yes	no	yes	no	5	0
ai.onnx	If	yes	yes	yes	yes	3	2
ai.onnx	ImageDecoder	yes	yes	yes	no	9	0
ai.onnx	InstanceNormalization	yes	no	yes	no	2	0
ai.onnx	IsInf	yes	no	yes	no	4	0
ai.onnx	IsNaN	yes	no	yes	no	2	0
ai.onnx	LRN	yes	no	yes	no	2	0
ai.onnx	LSTM	yes	yes	yes	no	4	0
ai.onnx	LayerNormalization	yes	no	yes	no	19	0
ai.onnx	LeakyRelu	yes	no	yes	no	3	0
ai.onnx	Less	yes	yes	yes	yes	26	4
ai.onnx	LessOrEqual	yes	no	yes	no	8	0
ai.onnx	Log	yes	no	yes	no	2	0
ai.onnx	LogSoftmax	yes	no	yes	no	7	0
ai.onnx	Loop	yes	no	yes	no	3	0
ai.onnx	LpNormalization	yes	no	yes	no	6	0
ai.onnx	LpPool	yes	no	yes	no	8	0
ai.onnx	MatMul	yes	yes	yes	no	7	0
ai.onnx	MatMulInteger	yes	no	yes	no	1	0
ai.onnx	Max	yes	no	yes	no	14	0
ai.onnx	MaxPool	yes	no	yes	no	19	0
ai.onnx	MaxRoiPool	yes	no	yes	no	0	0
ai.onnx	MaxUnpool	yes	no	yes	no	2	0
ai.onnx	Mean	yes	no	yes	no	3	0
ai.onnx	MeanVarianceNormalization	yes	no	no	no	1	0
ai.onnx	MelWeightMatrix	yes	no	yes	no	1	0
ai.onnx	Min	yes	no	yes	no	14	0
ai.onnx	Mish	yes	no	yes	no	1	0
ai.onnx	Mod	yes	yes	yes	no	13	0
ai.onnx	Mul	yes	yes	yes	no	27	5
ai.onnx	Multinomial	yes	no	yes	no	0	0
ai.onnx	Neg	yes	no	yes	no	2	0
ai.onnx	NegativeLogLikelihoodLoss	yes	no	yes	no	18	0
ai.onnx	NonMaxSuppression	yes	no	yes	no	10	0
ai.onnx	NonZero	yes	no	yes	no	1	0
ai.onnx	Not	yes	yes	yes	no	3	0
ai.onnx	OneHot	yes	no	yes	no	4	0
ai.onnx	Optional	yes	yes	yes	yes	0	1
ai.onnx	OptionalGetElement	yes	yes	yes	no	4	0
ai.onnx	OptionalHasElement	yes	yes	yes	no	7	0
ai.onnx	Or	yes	yes	yes	no	8	10
ai.onnx	PRelu	yes	no	yes	no	2	0
ai.onnx	Pad	yes	no	yes	no	6	0
ai.onnx	Pow	yes	yes	yes	no	12	0
ai.onnx	QLinearConv	yes	no	yes	no	1	0
ai.onnx	QLinearMatMul	yes	no	yes	no	8	0
ai.onnx	QuantizeLinear	yes	yes	yes	yes	13	2
ai.onnx	RMSNormalization	yes	no	yes	no	19	0
ai.onnx	RNN	yes	yes	yes	no	4	0
ai.onnx	RandomNormal	yes	no	yes	no	0	0
ai.onnx	RandomNormalLike	yes	no	yes	no	0	0
ai.onnx	RandomUniform	yes	no	yes	no	0	0
ai.onnx	RandomUniformLike	yes	no	yes	no	3	0
ai.onnx	Range	yes	no	yes	no	2	0
ai.onnx	Reciprocal	yes	no	yes	no	2	0
ai.onnx	ReduceL1	yes	yes	yes	no	9	0
ai.onnx	ReduceL2	yes	yes	yes	no	9	0
ai.onnx	ReduceLogSum	yes	yes	yes	no	5	0
ai.onnx	ReduceLogSumExp	yes	yes	yes	no	9	0
ai.onnx	ReduceMax	yes	yes	yes	no	10	0
ai.onnx	ReduceMean	yes	yes	yes	no	8	0
ai.onnx	ReduceMin	yes	yes	yes	no	10	0
ai.onnx	ReduceProd	yes	yes	yes	no	9	0
ai.onnx	ReduceSum	yes	yes	yes	yes	17	3
ai.onnx	ReduceSumSquare	yes	yes	yes	no	9	0
ai.onnx	RegexFullMatch	yes	yes	yes	no	3	0
ai.onnx	Relu	yes	no	yes	no	1	0
ai.onnx	Reshape	yes	no	yes	yes	10	0
ai.onnx	Resize	yes	no	yes	no	39	0
ai.onnx	ReverseSequence	yes	no	yes	no	2	0
ai.onnx	RoiAlign	yes	yes	yes	yes	3	2
ai.onnx	RotaryEmbedding	yes	no	yes	no	8	0
ai.onnx	Round	yes	no	yes	no	1	0
ai.onnx	STFT	yes	no	yes	no	2	0
ai.onnx	Scan	yes	no	yes	no	2	0
ai.onnx	Scatter	yes	no	yes	no	2	0
ai.onnx	ScatterElements	yes	no	yes	no	6	0
ai.onnx	ScatterND	yes	no	yes	no	5	0
ai.onnx	Selu	yes	no	yes	no	3	0
ai.onnx	SequenceAt	yes	yes	yes	no	0	0
ai.onnx	SequenceConstruct	yes	yes	yes	yes	0	2
ai.onnx	SequenceEmpty	yes	yes	yes	no	0	0
ai.onnx	SequenceErase	yes	yes	yes	no	0	0
ai.onnx	SequenceInsert	yes	yes	yes	no	2	0
ai.onnx	SequenceLength	yes	yes	yes	no	6	0
ai.onnx	SequenceMap	yes	yes	yes	no	6	0
ai.onnx	Shape	yes	no	yes	no	73	0
ai.onnx	Shrink	yes	no	yes	no	2	0
ai.onnx	Sigmoid	yes	no	yes	no	2	0
ai.onnx	Sign	yes	no	yes	no	1	0
ai.onnx	Sin	yes	yes	yes	no	2	0
ai.onnx	Sinh	yes	yes	yes	no	2	0
ai.onnx	Size	yes	no	yes	no	2	0
ai.onnx	Slice	yes	no	yes	no	8	0
ai.onnx	Softmax	yes	no	yes	no	7	0
ai.onnx	SoftmaxCrossEntropyLoss	yes	no	yes	no	34	0
ai.onnx	Softplus	yes	no	yes	no	3	0
ai.onnx	Softsign	yes	no	yes	no	2	0
ai.onnx	SpaceToDepth	yes	no	yes	no	2	0
ai.onnx	Split	yes	no	yes	no	16	0
ai.onnx	SplitToSequence	yes	yes	yes	no	3	0
ai.onnx	Sqrt	yes	no	yes	no	2	0
ai.onnx	Squeeze	yes	no	yes	no	2	0
ai.onnx	StringConcat	yes	yes	yes	yes	5	5
ai.onnx	StringNormalizer	yes	yes	yes	no	6	0
ai.onnx	StringSplit	yes	yes	yes	no	6	0
ai.onnx	Sub	yes	yes	yes	no	11	5
ai.onnx	Sum	yes	no	yes	no	3	0
ai.onnx	Swish	yes	no	yes	no	1	0
ai.onnx	Tan	yes	no	yes	no	2	0
ai.onnx	Tanh	yes	no	yes	no	2	0
ai.onnx	TensorScatter	yes	no	yes	no	3	0
ai.onnx	TfIdfVectorizer	yes	no	yes	no	7	0
ai.onnx	ThresholdedRelu	yes	no	yes	no	3	0
ai.onnx	Tile	yes	no	yes	no	2	0
ai.onnx	TopK	yes	no	yes	no	7	0
ai.onnx	Transpose	yes	no	yes	no	7	0
ai.onnx	Trilu	yes	no	yes	no	18	0
ai.onnx	Unique	yes	no	yes	no	6	0
ai.onnx	Unsqueeze	yes	no	yes	no	7	0
ai.onnx	Upsample	yes	no	yes	no	1	0
ai.onnx	Where	yes	no	yes	no	2	0
ai.onnx	Xor	yes	yes	yes	no	8	10
ai.onnx.ml	ArrayFeatureExtractor	yes	no	yes	no	1	0
ai.onnx.ml	Binarizer	yes	yes	yes	yes	1	2
ai.onnx.ml	CastMap	yes	no	yes	no	0	0
ai.onnx.ml	CategoryMapper	yes	no	yes	no	0	0
ai.onnx.ml	DictVectorizer	yes	no	yes	no	0	0
ai.onnx.ml	FeatureVectorizer	yes	no	no	no	0	0
ai.onnx.ml	Imputer	yes	no	no	no	0	0
ai.onnx.ml	LabelEncoder	yes	yes	yes	yes	4	2
ai.onnx.ml	LinearClassifier	yes	no	yes	no	0	0
ai.onnx.ml	LinearRegressor	yes	no	no	no	0	0
ai.onnx.ml	Normalizer	yes	no	no	no	0	0
ai.onnx.ml	OneHotEncoder	yes	no	yes	no	0	0
ai.onnx.ml	SVMClassifier	yes	no	yes	no	0	0
ai.onnx.ml	SVMRegressor	yes	no	no	no	0	0
ai.onnx.ml	Scaler	yes	no	no	no	0	0
ai.onnx.ml	TreeEnsemble	yes	yes	yes	no	2	0
ai.onnx.ml	TreeEnsembleClassifier	yes	yes	no	no	0	0
ai.onnx.ml	TreeEnsembleRegressor	yes	yes	no	no	0	0
ai.onnx.ml	ZipMap	yes	yes	yes	no	0	0
ai.onnx.preview	FlexAttention	no	yes	no	yes	0	2
ai.onnx.preview.training	Adagrad	yes	no	yes	no	2	0
ai.onnx.preview.training	Adam	yes	yes	yes	yes	2	4
ai.onnx.preview.training	Gradient	yes	yes	no	no	0	0
ai.onnx.preview.training	Momentum	yes	no	yes	no	3	0

---

See also

• Using backend tests to evaluate a runtime

• Backend test-case coverage