v0.1.0

ONNXNaiveNASflux v0.1.0

Closed issues:

• Add validation when serializing (#5)
• Verify exported models with other frameworks (#9)
• "ERROR: Use broadcasting (relu.(x))" when importing resnet18 (#17)
• Why does this depend on JuMP? (#19)
• Unsatisfiable requirements error on install in Julia 1.5.2 related to CuArrays (#25)
• Examples in Julia 1.5.2 (#26)
• Dimension errors in feedforward network serialization (#29)
• wrapfrom uses incorrect params signature (#31)
• minor ...documentation change for adding package from url... (#33)
• Broken CI (#34)
• type TensorProto has no field __protobuf_jl_internal_defaultset (#37)
• Failing to serialize flatten (#39)
• Potential issue with serialized data for conv layers (#41)
• Models with missing input shapes might be non-compliant (#42)
• Improve shape inference of Chains (#43)

Merged pull requests:

• Add force flag to prevent artifact exists error (#1) (@DrChainsaw)
• Serialize (#2) (@DrChainsaw)
• Add possibility to change how vertices are created (#3) (@DrChainsaw)
• Add RNN ser-deser (#4) (@DrChainsaw)
• Validate (#6) (@DrChainsaw)
• Add throttling and retry mechanism for artifact download (#7) (@DrChainsaw)
• Add testing with ONNX runtime as reference (#8) (@DrChainsaw)
• Sizeprop (#10) (@DrChainsaw)
• Resizeconstraint (#11) (@DrChainsaw)
• Add GlobalMaxPool (#13) (@DrChainsaw)
• Add support for Constant and Mul (#15) (@DrChainsaw)
• Fix typo (#16) (@DrChainsaw)
• Fix activation functions not being broadcasted (#18) (@DrChainsaw)
• Use Float32 when testing flux layers (#20) (@DrChainsaw)
• Set upper bound for NaiveNASflux (#21) (@DrChainsaw)
• Flux 0.11 compatibility (#22) (@DrChainsaw)
• Change CI to julia 1.5 (#23) (@DrChainsaw)
• Replace ONNX with BaseOnnx (#24) (@DrChainsaw)
• Some improvements to input shape validation and inference (#30) (@DrChainsaw)
• Fix instances where params was incorrectly called when merging ops (#32) (@DrChainsaw)
• Change to Condas builtin pip API to install onnxruntime (#35) (@DrChainsaw)
• move test dependencies to test/Project.toml (#36) (@DrChainsaw)
• Add 0.9.0 compat bound for ProtoBuf (#38) (@DrChainsaw)
• Add deserialization for Flux.flatten (#40) (@DrChainsaw)
• Serialize missing input shape using missing TensorShapeProto (#44) (@DrChainsaw)
• Improve shape inference for Chain (#45) (@DrChainsaw)
• set ci to 1.6 (#48) (@DrChainsaw)
• Flux0.12 (#49) (@DrChainsaw)
• Up compat for ProtoBuf to 0.11.1 (#50) (@DrChainsaw)
• Add let-block to prevent boxing of constants (#51) (@DrChainsaw)
• Rename package to ONNXNaiveNASflux (#52) (@DrChainsaw)