onnx_ort_flatbuffers.h

namespace onnx_light

Alias that makes onnx-light headers compatible with code that references ONNX_LIGHT_NAMESPACE (the macro used in the standard onnx package).

Set to ONNX_LIGHT_NAMESPACE so both names resolve to the same namespace.

Symbol-visibility attribute for the public onnx-light C++ API.

Maps the upstream compatibility macro to onnx-light’s explicit proto ABI annotation. This keeps declarations from vendored ONNX headers visible when lib_onnx_proto uses hidden visibility by default.

Namespace alias so that ONNX C++ code (and consumers such as onnxruntime) that refers to the literal onnx namespace — rather than the ONNX_NAMESPACE macro — resolves to the onnx-light namespace. The standard onnx package lives in namespace onnx; onnx-light uses onnx_light (via ONNX_LIGHT_NAMESPACE), so this alias keeps onnx-light a true drop-in. It is only introduced when the onnx-light namespace differs from onnx.

Functions

ONNX_LIGHT_PROTO_API std::string SerializeModelToOrtFlatbuffers (const ModelProto &model, const SerializeOptions &options)

Serializes a model into a native ORTM FlatBuffer.

Emits ORT format version 4, which full ONNX Runtime builds upgrade using their operator schemas. Minimal ONNX Runtime builds are not supported. Tensor data is embedded; external tensors must already have their payload loaded. Aligns numeric tensor payload offsets to at least eight bytes, or to the larger power-of-two alignment requested by the serialization options. Applies serialization callbacks without changing the supplied model. Infers intermediate element types for common standard operators; other operators require explicit graph value_info. Normalizes Constant nodes into initializers, as required by the ORT loader.

Throws std::invalid_argument for unsupported constructs or invalid input and std::length_error when the output exceeds the configured limit or the FlatBuffers signed 32-bit size limit. Model-local functions, optional/sparse/ opaque types, complex tensors, sparse initializers/attributes, GRAPHS attributes, and encoded initializers are unsupported. Tensor element types newer than FLOAT8E5M2FNUZ (20), including UINT4, INT4, FLOAT4E2M1, FLOAT8E8M0, UINT2, and INT2, are unsupported. Preserves model-level metadata_props but rejects graph, node, tensor, and value-info metadata_props, including in nested graphs and tensor attributes. Resolves formal input counts from an opset-versioned schema snapshot and rejects operators without a matching input schema, including unknown custom operators.

Returns: The complete FlatBuffer, including its ORTM file identifier.

ONNX_LIGHT_PROTO_API bool SerializeModelToOrtFlatbuffers (const ModelProto &model, std::string &output, const SerializeOptions &options)

Serializes a model into an output string.

Returns false and clears the output when the configured size limit is exceeded. Throws for other invalid or unsupported input.

ONNX_LIGHT_PROTO_API void ParseModelFromOrtFlatbuffers (ModelProto &model, utils::BinaryStream &stream, ParseOptions &options)

Parses an ORTM FlatBuffer into a model, replacing it only after successful decoding.

Accepts ORT versions 4, 5 and 6, including shared forward/backward vtables. Copies tensor payloads into owned, optionally aligned storage even when no_copy is requested. Decoding is serial regardless of num_threads; no external I/O scheduling or tracing is performed. Enforces recursion and tensor allocation limits before materializing fields. skip_raw_data omits numeric raw payloads; string tensor values remain available. Applies raw-data and node callbacks after graph reconstruction.

Rejects external tensor offsets/weights streams, fused nodes, saved runtime optimizations, unsupported schema extensions, and external-I/O policy options. Kernel resolver and placement metadata are checked but not retained in ONNX. Custom-domain operators are retained and require matching runtime kernels. Throws RuntimeError-compatible exceptions for malformed input and ParseLimitExceeded for configured resource limits.

namespace utils