onnx without protobuf
// readme
onnx-light
See also the ONNX roadmap for the upstream project’s direction and priorities.
Note:
onnx-lightstarted from the upstream ONNX pull request onnx/onnx#7208, which is the initial code base from which this project diverged.
onnx without protobuf
- ONNX Files larger than 2 GB (protobuf is limited to 2Gb)
- Parallel loading and saving: significantly faster compared to the single-threaded path
- Zero-copy parsing – creates the ModelProto without any tensor copy
- Aligned external tensor offsets – external tensor data can be written with explicit offset alignment
- No serialize/parse round-trip for C++ tools – the Python
ModelProtois the C++ModelProto - Supports protobuf (onnx) and flatbuffers (onnxruntime) format.
Modular C++ libraries
The C++ code is split into several small libraries so a downstream project can link only what it needs:
onnx_light::lib_onnx_proto– protobuf-compatible message types, parser / serializer, external data, optional…