mirror of
https://github.com/saymrwulf/onnxruntime.git
synced 2026-06-24 02:47:54 +00:00
218 lines
6.6 KiB
Text
218 lines
6.6 KiB
Text
{
|
|
"cells": [
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": 1,
|
|
"metadata": {},
|
|
"outputs": [],
|
|
"source": [
|
|
"import onnx\n",
|
|
"from onnx import helper, shape_inference\n",
|
|
"from onnx import TensorProto\n",
|
|
"import onnx.optimizer"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "markdown",
|
|
"metadata": {},
|
|
"source": [
|
|
"# Building a FC model for MNIST"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": 2,
|
|
"metadata": {},
|
|
"outputs": [],
|
|
"source": [
|
|
"import math\n",
|
|
"import numpy as np\n",
|
|
"import scipy.stats as stats\n",
|
|
"\n",
|
|
"def truncated_normal(dims): \n",
|
|
" size = 1\n",
|
|
" for dim in dims:\n",
|
|
" size *= dim\n",
|
|
"\n",
|
|
" mu, stddev = 0, 1/math.sqrt(size)\n",
|
|
" lower, upper = -2 * stddev, 2 * stddev\n",
|
|
" X = stats.truncnorm( (lower - mu) / stddev, (upper - mu) / stddev, loc = mu, scale = stddev)\n",
|
|
"\n",
|
|
" return X.rvs(size).tolist()\n",
|
|
"\n",
|
|
" \n",
|
|
"def zeros(dim):\n",
|
|
" return [0] * dim[0]\n",
|
|
" "
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": 3,
|
|
"metadata": {},
|
|
"outputs": [],
|
|
"source": [
|
|
"batch_size = -1\n",
|
|
"\n",
|
|
"W1_dims = [8, 1, 5, 5]\n",
|
|
"W2_dims = [16, 8, 5, 5]\n",
|
|
"W3_dims = [256, 10]\n",
|
|
"\n",
|
|
"W1 = helper.make_tensor(name=\"W1\", data_type=onnx.TensorProto.FLOAT, dims=W1_dims, vals=truncated_normal(W1_dims))\n",
|
|
"W2 = helper.make_tensor(name=\"W2\", data_type=onnx.TensorProto.FLOAT, dims=W2_dims, vals=truncated_normal(W2_dims))\n",
|
|
"W3 = helper.make_tensor(name=\"W3\", data_type=onnx.TensorProto.FLOAT, dims=W3_dims, vals=truncated_normal(W3_dims))\n",
|
|
"\n",
|
|
"B1_dims = [8]\n",
|
|
"B2_dims = [16]\n",
|
|
"B3_dims = [10]\n",
|
|
"\n",
|
|
"B1 = helper.make_tensor(name=\"B1\", data_type=onnx.TensorProto.FLOAT, dims=B1_dims, vals=zeros(B1_dims))\n",
|
|
"B2 = helper.make_tensor(name=\"B2\", data_type=onnx.TensorProto.FLOAT, dims=B2_dims, vals=zeros(B2_dims))\n",
|
|
"B3 = helper.make_tensor(name=\"B3\", data_type=onnx.TensorProto.FLOAT, dims=B3_dims, vals=zeros(B3_dims))\n",
|
|
"\n",
|
|
"\n",
|
|
"shape = helper.make_tensor(name=\"shape\", data_type=onnx.TensorProto.INT64, dims=[2], vals=[batch_size, 256])"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": 4,
|
|
"metadata": {},
|
|
"outputs": [],
|
|
"source": [
|
|
"node1 = helper.make_node('Conv', inputs=['X', 'W1', 'B1'], outputs=['T1'], kernel_shape=[5,5], strides=[1,1], pads=[2,2,2,2])\n",
|
|
"node2 = helper.make_node('Relu', inputs=['T1'], outputs=['T2'])\n",
|
|
"node3 = helper.make_node('MaxPool', inputs=['T2'], outputs=['T3'], kernel_shape=[2,2], strides=[2,2])\n",
|
|
"\n",
|
|
"node4 = helper.make_node('Conv', inputs=['T3', 'W2', 'B2'], outputs=['T4'], kernel_shape=[5,5], strides=[1,1], pads=[2,2,2,2])\n",
|
|
"node5 = helper.make_node('Relu', inputs=['T4'], outputs=['T5'])\n",
|
|
"node6 = helper.make_node('MaxPool', inputs=['T5'], outputs=['T6'], kernel_shape=[3,3], strides=[3,3])\n",
|
|
"\n",
|
|
"node7 = helper.make_node('Reshape', inputs=['T6', 'shape'], outputs=['T7'])\n",
|
|
"\n",
|
|
"node8 = helper.make_node('Gemm', inputs=['T7', 'W3', 'B3'], outputs=['predictions'])\n",
|
|
"\n",
|
|
"graph = helper.make_graph(\n",
|
|
" [node1, node2, node3, node4, node5, node6, node7, node8],\n",
|
|
" 'mnist_conv',\n",
|
|
" [helper.make_tensor_value_info('X', TensorProto.FLOAT, ([batch_size, 1, 28, 28])),\n",
|
|
" helper.make_tensor_value_info('W1', TensorProto.FLOAT, W1_dims),\n",
|
|
" helper.make_tensor_value_info('W2', TensorProto.FLOAT, W2_dims),\n",
|
|
" helper.make_tensor_value_info('W3', TensorProto.FLOAT, W3_dims),\n",
|
|
" helper.make_tensor_value_info('B1', TensorProto.FLOAT, B1_dims),\n",
|
|
" helper.make_tensor_value_info('B2', TensorProto.FLOAT, B2_dims),\n",
|
|
" helper.make_tensor_value_info('B3', TensorProto.FLOAT, B3_dims),\n",
|
|
" helper.make_tensor_value_info('shape', TensorProto.INT64, [2]),\n",
|
|
" ],\n",
|
|
" [helper.make_tensor_value_info('predictions', TensorProto.FLOAT, ([batch_size, 10]))],\n",
|
|
" [W1, W2, W3, B1, B2, B3, shape]\n",
|
|
")\n",
|
|
"original_model = helper.make_model(graph, producer_name='onnx-examples')"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": 5,
|
|
"metadata": {},
|
|
"outputs": [],
|
|
"source": [
|
|
"onnx.checker.check_model(original_model)"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": 6,
|
|
"metadata": {},
|
|
"outputs": [],
|
|
"source": [
|
|
"inferred_model = shape_inference.infer_shapes(original_model)\n",
|
|
"onnx.save_model(inferred_model, \"mnist_conv_batch_unknown.onnx\")"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "markdown",
|
|
"metadata": {},
|
|
"source": [
|
|
"# Inferencing session"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": 19,
|
|
"metadata": {},
|
|
"outputs": [],
|
|
"source": [
|
|
"import onnxruntime as lotus\n",
|
|
"import numpy as np"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": 20,
|
|
"metadata": {},
|
|
"outputs": [],
|
|
"source": [
|
|
"sess = lotus.InferenceSession('mnist_conv_batch.onnx', None)"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": 25,
|
|
"metadata": {},
|
|
"outputs": [
|
|
{
|
|
"name": "stdout",
|
|
"output_type": "stream",
|
|
"text": [
|
|
"[[ 0.00177374 -0.00565182 -0.00091211 -0.00833805 -0.00262394 0.00868058\n",
|
|
" -0.00174895 -0.00810127 0.01124816 0.01306908]\n",
|
|
" [ 0.00312558 -0.001757 -0.00442985 -0.00995069 -0.00409109 0.00804279\n",
|
|
" -0.00279926 -0.00883552 0.01240066 0.01041252]\n",
|
|
" [ 0.00396657 -0.00412453 -0.00415219 -0.01012361 -0.00570025 0.00737222\n",
|
|
" -0.00099745 -0.00989306 0.01327983 0.0105867 ]\n",
|
|
" [ 0.00241887 -0.00515954 -0.00304933 -0.01251889 -0.00433843 0.0081604\n",
|
|
" -0.00310066 -0.00968494 0.00971858 0.01077508]]\n"
|
|
]
|
|
}
|
|
],
|
|
"source": [
|
|
"X_dims = [4, 1, 28, 28]\n",
|
|
"\n",
|
|
"data = np.random.uniform(size=X_dims).astype(np.float32)\n",
|
|
"\n",
|
|
"result = sess.run(['predictions'], {'X': data})\n",
|
|
"\n",
|
|
"print(result[0])\n",
|
|
"\n"
|
|
]
|
|
},
|
|
{
|
|
"cell_type": "code",
|
|
"execution_count": null,
|
|
"metadata": {},
|
|
"outputs": [],
|
|
"source": []
|
|
}
|
|
],
|
|
"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": "3.6.5"
|
|
}
|
|
},
|
|
"nbformat": 4,
|
|
"nbformat_minor": 2
|
|
}
|