{
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "\n# Model Local Functions\n\nA model in ONNX may contain model-local functions. When converting an *onnxscript*\nfunction to a ModelProto, the default behavior is to include function-definitions\nfor all transitively called function-ops as model-local functions in the generated\nmodel (for which an *onnxscript* function definition has been seen). Callers can\noverride this behavior by explicitly providing the list of FunctionProtos to be\nincluded in the generated model.\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "First, let us define an ONNXScript function that calls other ONNXScript functions.\n\n"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "collapsed": false
      },
      "outputs": [],
      "source": [
        "import onnx\n\nfrom onnxscript import FLOAT, script\nfrom onnxscript import opset15 as op\nfrom onnxscript.values import Opset\n\n# A dummy opset used for model-local functions\nlocal = Opset(\"local\", 1)\n\n\n@script(local, default_opset=op)\ndef diff_square(x, y):\n    diff = x - y\n    return diff * diff\n\n\n@script(local)\ndef sum(z):\n    return op.ReduceSum(z, keepdims=1)\n\n\n@script()\ndef l2norm(x: FLOAT[\"N\"], y: FLOAT[\"N\"]) -> FLOAT[1]:  # noqa: F821\n    return op.Sqrt(sum(diff_square(x, y)))"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "Let's see what the generated model looks like by default:\n\n"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "collapsed": false
      },
      "outputs": [],
      "source": [
        "model = l2norm.to_model_proto()\nprint(onnx.printer.to_text(model))"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "Let's now explicitly specify which functions to include.\nFirst, generate a model with no model-local functions:\n\n"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "collapsed": false
      },
      "outputs": [],
      "source": [
        "model = l2norm.to_model_proto(functions=[])\nprint(onnx.printer.to_text(model))"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {},
      "source": [
        "Now, generate a model with one model-local function:\n\n"
      ]
    },
    {
      "cell_type": "code",
      "execution_count": null,
      "metadata": {
        "collapsed": false
      },
      "outputs": [],
      "source": [
        "model = l2norm.to_model_proto(functions=[sum])\nprint(onnx.printer.to_text(model))"
      ]
    }
  ],
  "metadata": {
    "kernelspec": {
      "display_name": "Python 3",
      "language": "python",
      "name": "python3"
    },
    "language_info": {
      "codemirror_mode": {
        "name": "ipython",
        "version": 3
      },
      "file_extension": ".py",
      "mimetype": "text/x-python",
      "name": "python",
      "nbconvert_exporter": "python",
      "pygments_lexer": "ipython3",
      "version": "3.10.15"
    }
  },
  "nbformat": 4,
  "nbformat_minor": 0
}