This article is translated from the English document Quick Start Tutorial for Compiling Deep Learning Models. The authors are Yao Wang, Truman Tian. More TVM Chinese documents can be accessed → TVM Chinese Station
This example shows how to build a neural network with the Relay Python front end and generate a runtime library for an NVIDIA GPU with TVM. Note that building TVM requires CUDA and LLVM to be enabled.
Overview of TVM Supported Hardware Backends
The image below shows the currently supported hardware backends for TVM: 
CUDA and LLVM will be selected as target backends for this tutorial. First, import Relay and TVM.
import numpy as np
from tvm import relay
from tvm.relay import testing
import tvm
from tvm import te
from tvm.contrib import graph_executor
import tvm.testing
Define a neural network in Relay
First, define a neural network with a Relay Python frontend. For simplicity, we use the predefined resnet-18 network in Relay. Parameters are initialized with the Xavier initializer. Relay also supports other model formats such as MXNet, CoreML, ONNX, and Tensorflow.
This tutorial assumes inference on our device with a batch size of 1. The input image is an RGB color image of size 224*224. Call tvm.relay.expr.TupleWrapper.astext() to view the network structure.
batch_size = 1
num_class = 1000
image_shape = (3, 224, 224)
data_shape = (batch_size,) + image_shape
out_shape = (batch_size, num_class)
mod, params = relay.testing.resnet.get_workload(
num_layers=18, batch_size=batch_size, image_shape=image_shape
)
# 想显示元数据则设置 show_meta_data=True
print(mod.astext(show_meta_data=False))
Visit mod.astext in the code for more information
Output result:
#[version = "0.0.5"]
def @main(%data: Tensor[(1, 3, 224, 224), float32] /* ty=Tensor[(1, 3, 224, 224), float32] */, %bn_data_gamma: Tensor[(3), float32] /* ty=Tensor[(3), float32] */, %bn_data_beta: Tensor[(3), float32] /* ty=Tensor[(3), float32] */, %bn_data_moving_mean: Tensor[(3), float32] /* ty=Tensor[(3), float32] */, %bn_data_moving_var: Tensor[(3), float32] /* ty=Tensor[(3), float32] */, %conv0_weight: Tensor[(64, 3, 7, 7), float32] /* ty=Tensor[(64, 3, 7, 7), float32] */, %bn0_gamma: Tensor[(64), float32] /* ty=Tensor[(64), float32] */, %bn0_beta: Tensor[(64), float32] /* ty=Tensor[(64), float32] */, %bn0_moving_mean: Tensor[(64), float32] /* ty=Tensor[(64), float32] */, %bn0_moving_var: Tensor[(64), float32] /* ty=Tensor[(64), float32] */, %stage1_unit1_bn1_gamma: Tensor[(64), float32] /* ty=Tensor[(64), float32] */, %stage1_unit1_bn1_beta: Tensor[(64), float32] /* ty=Tensor[(64), float32] */, %stage1_unit1_bn1_moving_mean: Tensor[(64), float32] /* ty=Tensor[(64), float32] */, %stage1_unit1_bn1_moving_var: Tensor[(64), float32] /* ty=Tensor[(64), float32] */, %stage1_unit1_conv1_weight: Tensor[(64, 64, 3, 3), float32] /* ty=Tensor[(64, 64, 3, 3), float32] */, %stage1_unit1_bn2_gamma: Tensor[(64), float32] /* ty=Tensor[(64), float32] */, %stage1_unit1_bn2_beta: Tensor[(64), float32] /* ty=Tensor[(64), float32] */, %stage1_unit1_bn2_moving_mean: Tensor[(64), float32] /* ty=Tensor[(64), float32] */, %stage1_unit1_bn2_moving_var: Tensor[(64), float32] /* ty=Tensor[(64), float32] */, %stage1_unit1_conv2_weight: Tensor[(64, 64, 3, 3), float32] /* ty=Tensor[(64, 64, 3, 3), float32] */, %stage1_unit1_sc_weight: Tensor[(64, 64, 1, 1), float32] /* ty=Tensor[(64, 64, 1, 1), float32] */, %stage1_unit2_bn1_gamma: Tensor[(64), float32] /* ty=Tensor[(64), float32] */, %stage1_unit2_bn1_beta: Tensor[(64), float32] /* ty=Tensor[(64), float32] */, %stage1_unit2_bn1_moving_mean: Tensor[(64), float32] /* ty=Tensor[(64), float32] */, %stage1_unit2_bn1_moving_var: Tensor[(64), float32] /* ty=Tensor[(64), float32] */, %stage1_unit2_conv1_weight: Tensor[(64, 64, 3, 3), float32] /* ty=Tensor[(64, 64, 3, 3), float32] */, %stage1_unit2_bn2_gamma: Tensor[(64), float32] /* ty=Tensor[(64), float32] */, %stage1_unit2_bn2_beta: Tensor[(64), float32] /* ty=Tensor[(64), float32] */, %stage1_unit2_bn2_moving_mean: Tensor[(64), float32] /* ty=Tensor[(64), float32] */, %stage1_unit2_bn2_moving_var: Tensor[(64), float32] /* ty=Tensor[(64), float32] */, %stage1_unit2_conv2_weight: Tensor[(64, 64, 3, 3), float32] /* ty=Tensor[(64, 64, 3, 3), float32] */, %stage2_unit1_bn1_gamma: Tensor[(64), float32] /* ty=Tensor[(64), float32] */, %stage2_unit1_bn1_beta: Tensor[(64), float32] /* ty=Tensor[(64), float32] */, %stage2_unit1_bn1_moving_mean: Tensor[(64), float32] /* ty=Tensor[(64), float32] */, %stage2_unit1_bn1_moving_var: Tensor[(64), float32] /* ty=Tensor[(64), float32] */, %stage2_unit1_conv1_weight: Tensor[(128, 64, 3, 3), float32] /* ty=Tensor[(128, 64, 3, 3), float32] */, %stage2_unit1_bn2_gamma: Tensor[(128), float32] /* ty=Tensor[(128), float32] */, %stage2_unit1_bn2_beta: Tensor[(128), float32] /* ty=Tensor[(128), float32] */, %stage2_unit1_bn2_moving_mean: Tensor[(128), float32] /* ty=Tensor[(128), float32] */, %stage2_unit1_bn2_moving_var: Tensor[(128), float32] /* ty=Tensor[(128), float32] */, %stage2_unit1_conv2_weight: Tensor[(128, 128, 3, 3), float32] /* ty=Tensor[(128, 128, 3, 3), float32] */, %stage2_unit1_sc_weight: Tensor[(128, 64, 1, 1), float32] /* ty=Tensor[(128, 64, 1, 1), float32] */, %stage2_unit2_bn1_gamma: Tensor[(128), float32] /* ty=Tensor[(128), float32] */, %stage2_unit2_bn1_beta: Tensor[(128), float32] /* ty=Tensor[(128), float32] */, %stage2_unit2_bn1_moving_mean: Tensor[(128), float32] /* ty=Tensor[(128), float32] */, %stage2_unit2_bn1_moving_var: Tensor[(128), float32] /* ty=Tensor[(128), float32] */, %stage2_unit2_conv1_weight: Tensor[(128, 128, 3, 3), float32] /* ty=Tensor[(128, 128, 3, 3), float32] */, %stage2_unit2_bn2_gamma: Tensor[(128), float32] /* ty=Tensor[(128), float32] */, %stage2_unit2_bn2_beta: Tensor[(128), float32] /* ty=Tensor[(128), float32] */, %stage2_unit2_bn2_moving_mean: Tensor[(128), float32] /* ty=Tensor[(128), float32] */, %stage2_unit2_bn2_moving_var: Tensor[(128), float32] /* ty=Tensor[(128), float32] */, %stage2_unit2_conv2_weight: Tensor[(128, 128, 3, 3), float32] /* ty=Tensor[(128, 128, 3, 3), float32] */, %stage3_unit1_bn1_gamma: Tensor[(128), float32] /* ty=Tensor[(128), float32] */, %stage3_unit1_bn1_beta: Tensor[(128), float32] /* ty=Tensor[(128), float32] */, %stage3_unit1_bn1_moving_mean: Tensor[(128), float32] /* ty=Tensor[(128), float32] */, %stage3_unit1_bn1_moving_var: Tensor[(128), float32] /* ty=Tensor[(128), float32] */, %stage3_unit1_conv1_weight: Tensor[(256, 128, 3, 3), float32] /* ty=Tensor[(256, 128, 3, 3), float32] */, %stage3_unit1_bn2_gamma: Tensor[(256), float32] /* ty=Tensor[(256), float32] */, %stage3_unit1_bn2_beta: Tensor[(256), float32] /* ty=Tensor[(256), float32] */, %stage3_unit1_bn2_moving_mean: Tensor[(256), float32] /* ty=Tensor[(256), float32] */, %stage3_unit1_bn2_moving_var: Tensor[(256), float32] /* ty=Tensor[(256), float32] */, %stage3_unit1_conv2_weight: Tensor[(256, 256, 3, 3), float32] /* ty=Tensor[(256, 256, 3, 3), float32] */, %stage3_unit1_sc_weight: Tensor[(256, 128, 1, 1), float32] /* ty=Tensor[(256, 128, 1, 1), float32] */, %stage3_unit2_bn1_gamma: Tensor[(256), float32] /* ty=Tensor[(256), float32] */, %stage3_unit2_bn1_beta: Tensor[(256), float32] /* ty=Tensor[(256), float32] */, %stage3_unit2_bn1_moving_mean: Tensor[(256), float32] /* ty=Tensor[(256), float32] */, %stage3_unit2_bn1_moving_var: Tensor[(256), float32] /* ty=Tensor[(256), float32] */, %stage3_unit2_conv1_weight: Tensor[(256, 256, 3, 3), float32] /* ty=Tensor[(256, 256, 3, 3), float32] */, %stage3_unit2_bn2_gamma: Tensor[(256), float32] /* ty=Tensor[(256), float32] */, %stage3_unit2_bn2_beta: Tensor[(256), float32] /* ty=Tensor[(256), float32] */, %stage3_unit2_bn2_moving_mean: Tensor[(256), float32] /* ty=Tensor[(256), float32] */, %stage3_unit2_bn2_moving_var: Tensor[(256), float32] /* ty=Tensor[(256), float32] */, %stage3_unit2_conv2_weight: Tensor[(256, 256, 3, 3), float32] /* ty=Tensor[(256, 256, 3, 3), float32] */, %stage4_unit1_bn1_gamma: Tensor[(256), float32] /* ty=Tensor[(256), float32] */, %stage4_unit1_bn1_beta: Tensor[(256), float32] /* ty=Tensor[(256), float32] */, %stage4_unit1_bn1_moving_mean: Tensor[(256), float32] /* ty=Tensor[(256), float32] */, %stage4_unit1_bn1_moving_var: Tensor[(256), float32] /* ty=Tensor[(256), float32] */, %stage4_unit1_conv1_weight: Tensor[(512, 256, 3, 3), float32] /* ty=Tensor[(512, 256, 3, 3), float32] */, %stage4_unit1_bn2_gamma: Tensor[(512), float32] /* ty=Tensor[(512), float32] */, %stage4_unit1_bn2_beta: Tensor[(512), float32] /* ty=Tensor[(512), float32] */, %stage4_unit1_bn2_moving_mean: Tensor[(512), float32] /* ty=Tensor[(512), float32] */, %stage4_unit1_bn2_moving_var: Tensor[(512), float32] /* ty=Tensor[(512), float32] */, %stage4_unit1_conv2_weight: Tensor[(512, 512, 3, 3), float32] /* ty=Tensor[(512, 512, 3, 3), float32] */, %stage4_unit1_sc_weight: Tensor[(512, 256, 1, 1), float32] /* ty=Tensor[(512, 256, 1, 1), float32] */, %stage4_unit2_bn1_gamma: Tensor[(512), float32] /* ty=Tensor[(512), float32] */, %stage4_unit2_bn1_beta: Tensor[(512), float32] /* ty=Tensor[(512), float32] */, %stage4_unit2_bn1_moving_mean: Tensor[(512), float32] /* ty=Tensor[(512), float32] */, %stage4_unit2_bn1_moving_var: Tensor[(512), float32] /* ty=Tensor[(512), float32] */, %stage4_unit2_conv1_weight: Tensor[(512, 512, 3, 3), float32] /* ty=Tensor[(512, 512, 3, 3), float32] */, %stage4_unit2_bn2_gamma: Tensor[(512), float32] /* ty=Tensor[(512), float32] */, %stage4_unit2_bn2_beta: Tensor[(512), float32] /* ty=Tensor[(512), float32] */, %stage4_unit2_bn2_moving_mean: Tensor[(512), float32] /* ty=Tensor[(512), float32] */, %stage4_unit2_bn2_moving_var: Tensor[(512), float32] /* ty=Tensor[(512), float32] */, %stage4_unit2_conv2_weight: Tensor[(512, 512, 3, 3), float32] /* ty=Tensor[(512, 512, 3, 3), float32] */, %bn1_gamma: Tensor[(512), float32] /* ty=Tensor[(512), float32] */, %bn1_beta: Tensor[(512), float32] /* ty=Tensor[(512), float32] */, %bn1_moving_mean: Tensor[(512), float32] /* ty=Tensor[(512), float32] */, %bn1_moving_var: Tensor[(512), float32] /* ty=Tensor[(512), float32] */, %fc1_weight: Tensor[(1000, 512), float32] /* ty=Tensor[(1000, 512), float32] */, %fc1_bias: Tensor[(1000), float32] /* ty=Tensor[(1000), float32] */) -> Tensor[(1, 1000), float32] {
%0 = nn.batch_norm(%data, %bn_data_gamma, %bn_data_beta, %bn_data_moving_mean, %bn_data_moving_var, epsilon=2e-05f, scale=False) /* ty=(Tensor[(1, 3, 224, 224), float32], Tensor[(3), float32], Tensor[(3), float32]) */;
%1 = %0.0 /* ty=Tensor[(1, 3, 224, 224), float32] */;
%2 = nn.conv2d(%1, %conv0_weight, strides=[2, 2], padding=[3, 3, 3, 3], channels=64, kernel_size=[7, 7]) /* ty=Tensor[(1, 64, 112, 112), float32] */;
%3 = nn.batch_norm(%2, %bn0_gamma, %bn0_beta, %bn0_moving_mean, %bn0_moving_var, epsilon=2e-05f) /* ty=(Tensor[(1, 64, 112, 112), float32], Tensor[(64), float32], Tensor[(64), float32]) */;
%4 = %3.0 /* ty=Tensor[(1, 64, 112, 112), float32] */;
%5 = nn.relu(%4) /* ty=Tensor[(1, 64, 112, 112), float32] */;
%6 = nn.max_pool2d(%5, pool_size=[3, 3], strides=[2, 2], padding=[1, 1, 1, 1]) /* ty=Tensor[(1, 64, 56, 56), float32] */;
%7 = nn.batch_norm(%6, %stage1_unit1_bn1_gamma, %stage1_unit1_bn1_beta, %stage1_unit1_bn1_moving_mean, %stage1_unit1_bn1_moving_var, epsilon=2e-05f) /* ty=(Tensor[(1, 64, 56, 56), float32], Tensor[(64), float32], Tensor[(64), float32]) */;
%8 = %7.0 /* ty=Tensor[(1, 64, 56, 56), float32] */;
%9 = nn.relu(%8) /* ty=Tensor[(1, 64, 56, 56), float32] */;
%10 = nn.conv2d(%9, %stage1_unit1_conv1_weight, padding=[1, 1, 1, 1], channels=64, kernel_size=[3, 3]) /* ty=Tensor[(1, 64, 56, 56), float32] */;
%11 = nn.batch_norm(%10, %stage1_unit1_bn2_gamma, %stage1_unit1_bn2_beta, %stage1_unit1_bn2_moving_mean, %stage1_unit1_bn2_moving_var, epsilon=2e-05f) /* ty=(Tensor[(1, 64, 56, 56), float32], Tensor[(64), float32], Tensor[(64), float32]) */;
%12 = %11.0 /* ty=Tensor[(1, 64, 56, 56), float32] */;
%13 = nn.relu(%12) /* ty=Tensor[(1, 64, 56, 56), float32] */;
%14 = nn.conv2d(%13, %stage1_unit1_conv2_weight, padding=[1, 1, 1, 1], channels=64, kernel_size=[3, 3]) /* ty=Tensor[(1, 64, 56, 56), float32] */;
%15 = nn.conv2d(%9, %stage1_unit1_sc_weight, padding=[0, 0, 0, 0], channels=64, kernel_size=[1, 1]) /* ty=Tensor[(1, 64, 56, 56), float32] */;
%16 = add(%14, %15) /* ty=Tensor[(1, 64, 56, 56), float32] */;
%17 = nn.batch_norm(%16, %stage1_unit2_bn1_gamma, %stage1_unit2_bn1_beta, %stage1_unit2_bn1_moving_mean, %stage1_unit2_bn1_moving_var, epsilon=2e-05f) /* ty=(Tensor[(1, 64, 56, 56), float32], Tensor[(64), float32], Tensor[(64), float32]) */;
%18 = %17.0 /* ty=Tensor[(1, 64, 56, 56), float32] */;
%19 = nn.relu(%18) /* ty=Tensor[(1, 64, 56, 56), float32] */;
%20 = nn.conv2d(%19, %stage1_unit2_conv1_weight, padding=[1, 1, 1, 1], channels=64, kernel_size=[3, 3]) /* ty=Tensor[(1, 64, 56, 56), float32] */;
%21 = nn.batch_norm(%20, %stage1_unit2_bn2_gamma, %stage1_unit2_bn2_beta, %stage1_unit2_bn2_moving_mean, %stage1_unit2_bn2_moving_var, epsilon=2e-05f) /* ty=(Tensor[(1, 64, 56, 56), float32], Tensor[(64), float32], Tensor[(64), float32]) */;
%22 = %21.0 /* ty=Tensor[(1, 64, 56, 56), float32] */;
%23 = nn.relu(%22) /* ty=Tensor[(1, 64, 56, 56), float32] */;
%24 = nn.conv2d(%23, %stage1_unit2_conv2_weight, padding=[1, 1, 1, 1], channels=64, kernel_size=[3, 3]) /* ty=Tensor[(1, 64, 56, 56), float32] */;
%25 = add(%24, %16) /* ty=Tensor[(1, 64, 56, 56), float32] */;
%26 = nn.batch_norm(%25, %stage2_unit1_bn1_gamma, %stage2_unit1_bn1_beta, %stage2_unit1_bn1_moving_mean, %stage2_unit1_bn1_moving_var, epsilon=2e-05f) /* ty=(Tensor[(1, 64, 56, 56), float32], Tensor[(64), float32], Tensor[(64), float32]) */;
%27 = %26.0 /* ty=Tensor[(1, 64, 56, 56), float32] */;
%28 = nn.relu(%27) /* ty=Tensor[(1, 64, 56, 56), float32] */;
%29 = nn.conv2d(%28, %stage2_unit1_conv1_weight, strides=[2, 2], padding=[1, 1, 1, 1], channels=128, kernel_size=[3, 3]) /* ty=Tensor[(1, 128, 28, 28), float32] */;
%30 = nn.batch_norm(%29, %stage2_unit1_bn2_gamma, %stage2_unit1_bn2_beta, %stage2_unit1_bn2_moving_mean, %stage2_unit1_bn2_moving_var, epsilon=2e-05f) /* ty=(Tensor[(1, 128, 28, 28), float32], Tensor[(128), float32], Tensor[(128), float32]) */;
%31 = %30.0 /* ty=Tensor[(1, 128, 28, 28), float32] */;
%32 = nn.relu(%31) /* ty=Tensor[(1, 128, 28, 28), float32] */;
%33 = nn.conv2d(%32, %stage2_unit1_conv2_weight, padding=[1, 1, 1, 1], channels=128, kernel_size=[3, 3]) /* ty=Tensor[(1, 128, 28, 28), float32] */;
%34 = nn.conv2d(%28, %stage2_unit1_sc_weight, strides=[2, 2], padding=[0, 0, 0, 0], channels=128, kernel_size=[1, 1]) /* ty=Tensor[(1, 128, 28, 28), float32] */;
%35 = add(%33, %34) /* ty=Tensor[(1, 128, 28, 28), float32] */;
%36 = nn.batch_norm(%35, %stage2_unit2_bn1_gamma, %stage2_unit2_bn1_beta, %stage2_unit2_bn1_moving_mean, %stage2_unit2_bn1_moving_var, epsilon=2e-05f) /* ty=(Tensor[(1, 128, 28, 28), float32], Tensor[(128), float32], Tensor[(128), float32]) */;
%37 = %36.0 /* ty=Tensor[(1, 128, 28, 28), float32] */;
%38 = nn.relu(%37) /* ty=Tensor[(1, 128, 28, 28), float32] */;
%39 = nn.conv2d(%38, %stage2_unit2_conv1_weight, padding=[1, 1, 1, 1], channels=128, kernel_size=[3, 3]) /* ty=Tensor[(1, 128, 28, 28), float32] */;
%40 = nn.batch_norm(%39, %stage2_unit2_bn2_gamma, %stage2_unit2_bn2_beta, %stage2_unit2_bn2_moving_mean, %stage2_unit2_bn2_moving_var, epsilon=2e-05f) /* ty=(Tensor[(1, 128, 28, 28), float32], Tensor[(128), float32], Tensor[(128), float32]) */;
%41 = %40.0 /* ty=Tensor[(1, 128, 28, 28), float32] */;
%42 = nn.relu(%41) /* ty=Tensor[(1, 128, 28, 28), float32] */;
%43 = nn.conv2d(%42, %stage2_unit2_conv2_weight, padding=[1, 1, 1, 1], channels=128, kernel_size=[3, 3]) /* ty=Tensor[(1, 128, 28, 28), float32] */;
%44 = add(%43, %35) /* ty=Tensor[(1, 128, 28, 28), float32] */;
%45 = nn.batch_norm(%44, %stage3_unit1_bn1_gamma, %stage3_unit1_bn1_beta, %stage3_unit1_bn1_moving_mean, %stage3_unit1_bn1_moving_var, epsilon=2e-05f) /* ty=(Tensor[(1, 128, 28, 28), float32], Tensor[(128), float32], Tensor[(128), float32]) */;
%46 = %45.0 /* ty=Tensor[(1, 128, 28, 28), float32] */;
%47 = nn.relu(%46) /* ty=Tensor[(1, 128, 28, 28), float32] */;
%48 = nn.conv2d(%47, %stage3_unit1_conv1_weight, strides=[2, 2], padding=[1, 1, 1, 1], channels=256, kernel_size=[3, 3]) /* ty=Tensor[(1, 256, 14, 14), float32] */;
%49 = nn.batch_norm(%48, %stage3_unit1_bn2_gamma, %stage3_unit1_bn2_beta, %stage3_unit1_bn2_moving_mean, %stage3_unit1_bn2_moving_var, epsilon=2e-05f) /* ty=(Tensor[(1, 256, 14, 14), float32], Tensor[(256), float32], Tensor[(256), float32]) */;
%50 = %49.0 /* ty=Tensor[(1, 256, 14, 14), float32] */;
%51 = nn.relu(%50) /* ty=Tensor[(1, 256, 14, 14), float32] */;
%52 = nn.conv2d(%51, %stage3_unit1_conv2_weight, padding=[1, 1, 1, 1], channels=256, kernel_size=[3, 3]) /* ty=Tensor[(1, 256, 14, 14), float32] */;
%53 = nn.conv2d(%47, %stage3_unit1_sc_weight, strides=[2, 2], padding=[0, 0, 0, 0], channels=256, kernel_size=[1, 1]) /* ty=Tensor[(1, 256, 14, 14), float32] */;
%54 = add(%52, %53) /* ty=Tensor[(1, 256, 14, 14), float32] */;
%55 = nn.batch_norm(%54, %stage3_unit2_bn1_gamma, %stage3_unit2_bn1_beta, %stage3_unit2_bn1_moving_mean, %stage3_unit2_bn1_moving_var, epsilon=2e-05f) /* ty=(Tensor[(1, 256, 14, 14), float32], Tensor[(256), float32], Tensor[(256), float32]) */;
%56 = %55.0 /* ty=Tensor[(1, 256, 14, 14), float32] */;
%57 = nn.relu(%56) /* ty=Tensor[(1, 256, 14, 14), float32] */;
%58 = nn.conv2d(%57, %stage3_unit2_conv1_weight, padding=[1, 1, 1, 1], channels=256, kernel_size=[3, 3]) /* ty=Tensor[(1, 256, 14, 14), float32] */;
%59 = nn.batch_norm(%58, %stage3_unit2_bn2_gamma, %stage3_unit2_bn2_beta, %stage3_unit2_bn2_moving_mean, %stage3_unit2_bn2_moving_var, epsilon=2e-05f) /* ty=(Tensor[(1, 256, 14, 14), float32], Tensor[(256), float32], Tensor[(256), float32]) */;
%60 = %59.0 /* ty=Tensor[(1, 256, 14, 14), float32] */;
%61 = nn.relu(%60) /* ty=Tensor[(1, 256, 14, 14), float32] */;
%62 = nn.conv2d(%61, %stage3_unit2_conv2_weight, padding=[1, 1, 1, 1], channels=256, kernel_size=[3, 3]) /* ty=Tensor[(1, 256, 14, 14), float32] */;
%63 = add(%62, %54) /* ty=Tensor[(1, 256, 14, 14), float32] */;
%64 = nn.batch_norm(%63, %stage4_unit1_bn1_gamma, %stage4_unit1_bn1_beta, %stage4_unit1_bn1_moving_mean, %stage4_unit1_bn1_moving_var, epsilon=2e-05f) /* ty=(Tensor[(1, 256, 14, 14), float32], Tensor[(256), float32], Tensor[(256), float32]) */;
%65 = %64.0 /* ty=Tensor[(1, 256, 14, 14), float32] */;
%66 = nn.relu(%65) /* ty=Tensor[(1, 256, 14, 14), float32] */;
%67 = nn.conv2d(%66, %stage4_unit1_conv1_weight, strides=[2, 2], padding=[1, 1, 1, 1], channels=512, kernel_size=[3, 3]) /* ty=Tensor[(1, 512, 7, 7), float32] */;
%68 = nn.batch_norm(%67, %stage4_unit1_bn2_gamma, %stage4_unit1_bn2_beta, %stage4_unit1_bn2_moving_mean, %stage4_unit1_bn2_moving_var, epsilon=2e-05f) /* ty=(Tensor[(1, 512, 7, 7), float32], Tensor[(512), float32], Tensor[(512), float32]) */;
%69 = %68.0 /* ty=Tensor[(1, 512, 7, 7), float32] */;
%70 = nn.relu(%69) /* ty=Tensor[(1, 512, 7, 7), float32] */;
%71 = nn.conv2d(%70, %stage4_unit1_conv2_weight, padding=[1, 1, 1, 1], channels=512, kernel_size=[3, 3]) /* ty=Tensor[(1, 512, 7, 7), float32] */;
%72 = nn.conv2d(%66, %stage4_unit1_sc_weight, strides=[2, 2], padding=[0, 0, 0, 0], channels=512, kernel_size=[1, 1]) /* ty=Tensor[(1, 512, 7, 7), float32] */;
%73 = add(%71, %72) /* ty=Tensor[(1, 512, 7, 7), float32] */;
%74 = nn.batch_norm(%73, %stage4_unit2_bn1_gamma, %stage4_unit2_bn1_beta, %stage4_unit2_bn1_moving_mean, %stage4_unit2_bn1_moving_var, epsilon=2e-05f) /* ty=(Tensor[(1, 512, 7, 7), float32], Tensor[(512), float32], Tensor[(512), float32]) */;
%75 = %74.0 /* ty=Tensor[(1, 512, 7, 7), float32] */;
%76 = nn.relu(%75) /* ty=Tensor[(1, 512, 7, 7), float32] */;
%77 = nn.conv2d(%76, %stage4_unit2_conv1_weight, padding=[1, 1, 1, 1], channels=512, kernel_size=[3, 3]) /* ty=Tensor[(1, 512, 7, 7), float32] */;
%78 = nn.batch_norm(%77, %stage4_unit2_bn2_gamma, %stage4_unit2_bn2_beta, %stage4_unit2_bn2_moving_mean, %stage4_unit2_bn2_moving_var, epsilon=2e-05f) /* ty=(Tensor[(1, 512, 7, 7), float32], Tensor[(512), float32], Tensor[(512), float32]) */;
%79 = %78.0 /* ty=Tensor[(1, 512, 7, 7), float32] */;
%80 = nn.relu(%79) /* ty=Tensor[(1, 512, 7, 7), float32] */;
%81 = nn.conv2d(%80, %stage4_unit2_conv2_weight, padding=[1, 1, 1, 1], channels=512, kernel_size=[3, 3]) /* ty=Tensor[(1, 512, 7, 7), float32] */;
%82 = add(%81, %73) /* ty=Tensor[(1, 512, 7, 7), float32] */;
%83 = nn.batch_norm(%82, %bn1_gamma, %bn1_beta, %bn1_moving_mean, %bn1_moving_var, epsilon=2e-05f) /* ty=(Tensor[(1, 512, 7, 7), float32], Tensor[(512), float32], Tensor[(512), float32]) */;
%84 = %83.0 /* ty=Tensor[(1, 512, 7, 7), float32] */;
%85 = nn.relu(%84) /* ty=Tensor[(1, 512, 7, 7), float32] */;
%86 = nn.global_avg_pool2d(%85) /* ty=Tensor[(1, 512, 1, 1), float32] */;
%87 = nn.batch_flatten(%86) /* ty=Tensor[(1, 512), float32] */;
%88 = nn.dense(%87, %fc1_weight, units=1000) /* ty=Tensor[(1, 1000), float32] */;
%89 = nn.bias_add(%88, %fc1_bias, axis=-1) /* ty=Tensor[(1, 1000), float32] */;
nn.softmax(%89) /* ty=Tensor[(1, 1000), float32] */
}
compile
The next step is to compile the model with the Relay/TVM pipeline. The user can specify the optimization level for compilation (currently this value is 0 to 3). The optimization pass includes operator fusion, precomputation, layout transformation, etc.
relay.build() returns three components: an execution graph in JSON format, a TVM block library of functions to compile this execution graph on the target hardware, and the model's blobs parameter. During compilation, Relay performs graph-level optimization, while TVM performs tensor-level optimization, thus providing an optimized runtime module for model serving.
Compile for NVIDIA GPUs first. In this process, relay.build() first performs some graph-level optimization, such as pruning, fusion, etc., and then registers operators (that is, nodes of the optimized graph) to the implementation of TVM to generate tvm.module. To generate a module library, TVM first converts the high-level IR to the low-level native IR for the specified target backend, in this case CUDA. The resulting machine code will then serve as a module library.
opt_level = 3
target = tvm.target.cuda()
with tvm.transform.PassContext(opt_level=opt_level):
lib = relay.build(mod, target, params=params)
Output result:
/workspace/python/tvm/target/target.py:377: UserWarning: Try specifying cuda arch by adding 'arch=sm_xx' to your target.
warnings.warn("Try specifying cuda arch by adding 'arch=sm_xx' to your target.")
/workspace/python/tvm/driver/build_module.py:268: UserWarning: target_host parameter is going to be deprecated. Please pass in tvm.target.Target(target, host=target_host) instead.
"target_host parameter is going to be deprecated. "
Run build library
Create a graph executor and run the module on an NVIDIA GPU.
# create random input
dev = tvm.cuda()
data = np.random.uniform(-1, 1, size=data_shape).astype("float32")
# create module
module = graph_executor.GraphModule(lib["default"](dev))
# set input and parameters
module.set_input("data", data)
# run
module.run()
# get output
out = module.get_output(0, tvm.nd.empty(out_shape)).numpy()
# Print first 10 elements of output
print(out.flatten()[0:10])
Output result:
[0.00089283 0.00103331 0.0009094 0.00102275 0.00108751 0.00106737
0.00106262 0.00095838 0.00110792 0.00113151]
Save and load compiled modules
Computational graphs, libraries, and parameters can also be saved to files and then loaded in the deployment environment.
# 分别将计算图、库和参数保存到不同文件
from tvm.contrib import utils
temp = utils.tempdir()
path_lib = temp.relpath("deploy_lib.tar")
lib.export_library(path_lib)
print(temp.listdir())
Output result:
['deploy_lib.tar']
# 重新加载模块
loaded_lib = tvm.runtime.load_module(path_lib)
input_data = tvm.nd.array(data)
module = graph_executor.GraphModule(loaded_lib["default"](dev))
module.run(data=input_data)
out_deploy = module.get_output(0).numpy()
# 打印输出的前十个元素
print(out_deploy.flatten()[0:10])
# 检查来自部署模块的输出和原始输出是否一致
tvm.testing.assert_allclose(out_deploy, out, atol=1e-5)
Output result:
[0.00089283 0.00103331 0.0009094 0.00102275 0.00108751 0.00106737
0.00106262 0.00095838 0.00110792 0.00113151]
Download Python source code: relay_quick_start.py
Download Jupyter Notebook: relay_quick_start.ipynb