-m yobx print … print an onnx model on standard output

The command loads an onnx model and prints it on standard output in one of several formats.

Description

See yobx.helpers.onnx_helper.pretty_onnx() and yobx.xshape.BasicShapeBuilder.

    usage: print [-h] {dot,onnx-compact,mermaid,pretty,printer,raw,shape} input
    
    Prints the model on the standard output.
    
    positional arguments:
      {dot,onnx-compact,mermaid,pretty,printer,raw,shape}
                            Prints out a model on the standard output.
                            
                            dot          - converts the graph into dot
                            onnx-compact - translates the model into compact Python code
                            pretty       - an improved rendering
                            printer      - onnx.printer.to_text(...)
                            raw          - just prints the model with print(...)
                            shape        - prints every node with input and output shapes
                            
      input                 onnx model to load
    
    options:
      -h, --help            show this help message and exit
    
    To show a model.

Examples

Print a model in the default pretty format:

python -m yobx print pretty model.onnx

Print every node with its inferred input/output shapes:

python -m yobx print shape model.onnx

Print the raw protobuf representation:

python -m yobx print raw model.onnx

Print the onnx-native text representation:

python -m yobx print printer model.onnx

Dump the dot graph source:

python -m yobx print dot model.onnx

Output on a Dummy Model

The following block builds a small Add + Relu model on the fly and runs the command to show what the output actually looks like.

    --- pretty ---
    opset: domain='' version=18
    input: name='X' type=dtype('float32') shape=[2, 3]
    input: name='Y' type=dtype('float32') shape=[2, 3]
    Add(X, Y) -> added
      Relu(added) -> Z
    output: name='Z' type=dtype('float32') shape=[2, 3]
    
    --- printer ---
    <
       ir_version: 10,
       opset_import: ["" : 18]
    >
    add_relu (float[2,3] X, float[2,3] Y) => (float[2,3] Z) {
       added = Add (X, Y)
       Z = Relu (added)
    }