onnion-rt 0.8.0

run onnx with only numpy

onnion-rt

Note: This software includes the work that is distributed in the Apache License 2.0.

Installation

$ pip3 install onnion-rt

Usage

See tutorial.

Development Guide

$ poetry install

How to support new operators

1. Add onnion_runtime/xxx.py

• An onnx operator Xxx must correspond to a class Xxx.
• A class Xxx must implement __init__ and run methods.
• The parameters of the __init__ methods must be self, opset_version, and kwargs.
• The attributes of the operator must be passed through the kwargs of the __init__ method.
  • Get the required attributes by kwargs['attr_name'].
  • Get the optional attributes by kwargs.get('attr_name', default_value).
• The inputs of the operator must be passed through the arguments of the run method.
• The run method must return the list of np.array.

2. Add from .xxx import Xxx # noqa: F401 to onnion_runtime/__init__.py
3. Update "Supported Operators" in README.md
4. Add tests/test_xxx.py
5. Run tests poetry run pytest -v
6. Format and lint poetry run pysen run format && poetry run pysen run lint

Supported Operators

This runtime supports only below operators.

• Abs
• Acos
• Acosh
• Add
  • must be from opsetversion >= 7
• And
  • must be from opsetversion >= 7
• ArgMax
• ArgMin
• Asin
• Asinh
• Atan
• Atanh
• BatchNormalization
  • must be from opsetversion >= 9
• BitShift
• Cast
  • must be from opsetversion >= 6
• Ceil
• Celu
• Clip
• Compress
• Concat
• ConcatFromSequence
• Constant
• ConstantOfShape
• ConvTranspose
  • support 2d only
  • group should be 1
  • auto_pad should be "NOTSET" (default value)
• Cos
• Cosh
• DepthToSpace
• DequantizeLinear
• Det
• Div
  • must be from opsetversion >= 7
• Dropout
• DynamicQuantizeLinear
• Einsum
• Elu
• Equal
  • must be from opsetversion >= 7
• Erf
• Exp
• Expand
• EyeLike
• Flatten
• Floor
• Gather
• GatherElements
• GahterND
• Gemm
  • must be from opsetversion >= 7
• GlobalAveragePool
• GlobalMaxPool
• Greater
  • must be from opsetversion >= 7
• GreaterOrEqual
• HardSigmoid
• HardSwish
• Hardmax
• Identity
• If
• InstanceNormalization
• IsInf
• IsNaN
• LeakyRelu
• Less
  • must be from opsetversion >= 7
• LessOrEqual
• Log
• LogSoftmax
• Loop
• MatMul
• MatMulInteger
• Max
• Mean
• Min
• Mod
• Mul
  • must be from opsetversion >= 7
• Neg
• NegativeLogLikelihoodLoss
• NonMaxSuppression
• NonZero
• Not
• OneHot
• Or
  • must be from opsetversion >= 7
• PRelu
• Pad
• Pow
  • must be from opsetversion >= 7
• RandomNormal
• RandomNormalLike
• RandomUniform
• RandomUniformLike
• Range
• Reciprocal
• ReduceL1
• ReduceL2
• ReduceLogSum
• ReduceLogSumExp
• ReduceMax
• ReduceMean
• ReduceMin
• ReduceProd
• ReduceSum
• ReduceSumSquare
• Relu
• Reshape
• Round
• ScatterND
• Shape
• Sigmoid
• Slice
• Softmax
• Split
  • argument split must be specified
• Sqrt
• Squeeze
• Sub
  • must be from opsetversion >= 7
• Tile
  • must be from opsetversion >= 6
• TopK
• Transpose
• Unsqueeze
• Where