## ONNX v0.2.0

[Diff since v0.1.1](FluxML/ONNX.jl@v0.1.1...v0.2.0)


**Closed issues:**
- Register v0.1.1 (#23)
- models returned are generic functions, not Flux models (#29)
- Remove DataFlow.jl, use IRTools (#30)
- KeyError: key :TreeEnsembleRegressor not found (#38)
- ] test ONNX fails: Package Test not found in current path [but it is installed) (#40)
- Cannot load any ONNX models - 28 tests fail (#43)
- ERROR: KeyError: key :Gather not found (#52)
- current steps to load model to Flux with ONNX#master (#64)

**Merged pull requests:**
- Create .gitattributes (#44) (@avik-pal)
- Fresh start (#46) (@dfdx)
- small typo fix (#47) (@EMCP)
- Up ProtoBuf compat to 0.11.1 (#48) (@DrChainsaw)
- Import to Tape (#50) (@dfdx)
- Julia <> ONNX conversions + Conv (#51) (@dfdx)
- Replace PyCall with ONNXRunTime.jl in tests (#54) (@dfdx)
- Mul, Relu, MaxPool (#55) (@dfdx)
- Add Gemm (#56) (@dfdx)
- BatchNormalization (#57) (@dfdx)
- Global average pool (#58) (@dfdx)
- Resnet18 example (#59) (@dfdx)
- Gather (#63) (@dfdx)
- Adjust NNlib version to match that in Flux (#65) (@dfdx)
- Add Unsquezee operation (#66) (@dfdx)
- Add Slice (#67) (@dfdx)
- Represent array size with SVector (#68) (@dfdx)
- Add MatMul (#69) (@dfdx)

See github.com/FluxML/ONNX.jl/releases/tag/v0.1.1 for release notes

Final changes