Rule Reference

ONNX Doctor includes 64 rules across multiple providers (10 with auto-fix 🔧).

ONNX Rules

Code	Name	Severity	Fix	Target	Message
ONNX001	empty-graph-name	error	🔧	graph	Graph name of the root graph is empty.
ONNX002	empty-initializer-name	error	🔧	graph	Initializer name is empty.
ONNX003	unsorted-graph-nodes	error	🔧	graph	Graph nodes are not in topological order.
ONNX004	unknown-node-input	error		graph	Node input is not produced by a previous node or a graph input.
ONNX005	experimental-op	warning		graph	Node uses an experimental operator.
ONNX006	duplicate-value-name	error		graph	Multiple values share the same name within a graph.
ONNX007	graph-output-not-produced	error	🔧	graph	Graph output is not produced by any node in the graph.
ONNX008	graph-ssa-violation	error		graph	Value name is assigned multiple times, violating SSA form.
ONNX009	initializer-name-conflict	error		graph	Initializer name conflicts with a subgraph input name.
ONNX010	invalid-ir-version	error		model	Model ir_version is missing or invalid.
ONNX011	ir-version-too-new	warning		model	Model ir_version is newer than the checker supports.
ONNX012	duplicate-metadata-keys	error		model	Model has duplicate keys in metadata_props.
ONNX013	missing-default-opset	error		model	Model (IR version >= 3) is missing opset import for the default ONNX domain.
ONNX014	unexpected-opset-import	error		model	Model (IR version < 3) should not have opset_import specified.
ONNX015	missing-opset-for-domain	error		node	No opset imported for the node’s operator domain.
ONNX016	deprecated-op	warning		node	Node uses a deprecated operator.
ONNX017	unregistered-op	error		node	No operator schema found for this op type and opset version.
ONNX018	missing-value-type	info		value	Value has no type annotation.
ONNX019	undefined-value-dtype	error		value	Value’s tensor type has UNDEFINED dtype.
ONNX020	undefined-tensor-dtype	error		value	Tensor dtype is UNDEFINED.
ONNX021	external-tensor-absolute-path	error		value	External tensor location is an absolute path (must be relative).
ONNX022	external-tensor-empty-location	error		value	External tensor location is empty.
ONNX023	external-tensor-outside-model-dir	error		value	External tensor path points outside the model directory.
ONNX024	external-tensor-inaccessible	error		value	External tensor file does not exist or is not accessible.
ONNX025	external-tensor-not-file	error		value	External tensor path is not a regular file.
ONNX026	function-empty-name	error		function	Function name is empty.
ONNX027	function-missing-domain	error		function	Function has no domain (required for IR version >= 8).
ONNX028	function-duplicate-inputs	error		function	Function has duplicate input names.
ONNX029	function-duplicate-outputs	error		function	Function has duplicate output names.
ONNX030	function-duplicate-attributes	error		function	Function has duplicate attribute names.
ONNX031	unsorted-function-nodes	error	🔧	function	Function nodes are not in topological order.
ONNX032	function-ssa-violation	error		function	Function is not in SSA form — a value name is assigned multiple times.
ONNX033	function-opset-mismatch	error		function	Function opset import is incompatible with the model’s opset import.
ONNX034	graph-input-missing-type	error		graph	Graph input is missing type information.
ONNX035	graph-input-missing-shape	error		graph	Graph input is missing shape information.
ONNX036	graph-output-missing-type	error		graph	Graph output is missing type information.
ONNX037	graph-output-missing-shape	error		graph	Graph output is missing shape information.
ONNX038	subgraph-variable-shadowing	error		graph	Subgraph value name shadows a name from an outer scope.
ONNX039	ref-attr-outside-function	error		graph	Reference attribute can only appear in function nodes, not in the main graph or its subgraphs.
ONNX101	duplicate-graph-input	error		graph	Duplicate Value object in graph inputs.
ONNX102	duplicate-graph-output	error	🔧	graph	Duplicate Value object in graph outputs.
ONNX103	empty-value-name	warning	🔧	graph	Value has no name set.
ONNX104	initializer-missing-const-value	error		graph	Initializer Value has no const_value set.

ORT Rules

Code	Name	Severity	Fix	Target	Message
ORT001	operator-unsupported	error		node	Operator is not supported by the execution provider in ONNX Runtime.
ORT002	operator-version-unsupported	error		node	Operator at this opset version is not supported by the execution provider.
ORT003	node-type-inconsistent	error		node	Type constraint mismatch for operator inputs/outputs.
ORT004	type-unsupported	error		node	Data type is not supported by the execution provider for this operator.
ORT005	typestr-not-in-schema	error		node	Type string is not defined in the operator schema.

PB Rules

Code	Name	Severity	Fix	Target	Message
PB001	non-unique-initializer-name	error		graph	Tensor initializer names within a graph are not unique.
PB002	non-unique-sparse-initializer-name	error		graph	Sparse tensor initializer names are not unique.
PB003	attribute-multiple-value-fields	error		node	Attribute contains more than one value field.
PB004	attribute-ref-has-value	error		node	Attribute referring to a parent attribute has its own value field set.
PB005	tensor-data-field-mismatch	error		value	Tensor data field does not match the specified data type.
PB006	zero-element-tensor-has-data	error		value	A zero-element tensor should not contain data.
PB007	multi-value-field-tensor	error		value	Tensor with non-zero elements has multiple value fields set.
PB008	string-data-in-raw-data	error		value	STRING data stored in raw_data field.
PB009	externally-stored-tensor-has-data	error		value	Externally stored tensor has inline data fields set.
PB010	invalid-sparse-indices-size	error		value	Sparse tensor indices size does not match NNZ count.
PB011	sparse-indices-out-of-range	error		value	Sparse tensor indices are out of range.
PB012	unsorted-sparse-indices	error		value	Sparse tensor indices are not sorted.
PB013	sparse-indices-not-lexicographic	error		value	Sparse tensor indices are not in lexicographic order.

SIM Rules

Code	Name	Severity	Fix	Target	Message
SIM001	unused-functions	info	🔧	model	Model contains functions that are not referenced by any node.
SIM002	unused-opset-imports	info	🔧	model	Model imports opsets that are not used by any node.
SIM003	unused-node	info	🔧	node	Node outputs are not consumed.