1.1. Installation¶
Each section of this book is a Jupyter notebook. The easiest way to run them is clicking the COLAB button on the upper right of the HTML page, which will directly you to Google Colab with the corresponding notebook opened. Running the first code cell will connect to a host runtime and show the following warning message. You can click RUN ANYWAY to continue.
Fig. 1.1.1 Click RUN ANYWAY to run a section on Colab.¶
The reset of this section will go through how to set up a Python environment, Jupyter’s interactive notebooks, the relevant libraries, and the code needed to run the book you can run them on your machines.
1.1.1. Obtaining Source Codes¶
The source code package containing all notebooks is available at
http://tvm.d2l.ai/d2l-tvm.zip. Please download it and extract it into a
folder. For example, on Linux/macOS, if you have both wget and
unzip installed, you can do it through:
wget http://tvm.d2l.ai/d2l-tvm.zip
unzip d2l-tvm.zip -d d2l-tvm
1.1.2. Installing Running Environment¶
If you have both Python 3.5 or later and pip installed, the
easiest way to install the running environment is through pip. The
required packages are
d2ltvmfor all dependencies such as Jupyter and saved code blockstvm[Chen et al., 2018] for the deep learning compiler we are usingmxnetas the baseline in some chapters
First install d2ltvm:
pip install git+https://github.com/d2l-ai/d2l-tvm
Then compile tvm from source codes. TVM doesn’t have a pip package
because it highly depends on the libraries available on your system.
Please follow the instructions on
tvm.ai to install
tvm. The configuration in config.cmake this book requires at
least
set(USE_LLVM ON)
If you plan to run on Nvidia GPUs as well, you will also need to
set(USE_CUDA ON)
Also don’t forget to enable cython, which accelerates the
performance. You just need to run make cython in the TVM source
folder.
If luckily you are using Ubuntu with python-3.7, llvm-6.0 and
cuda-10.1 installed, you may use the pre-built library that is for
evaluating this book:
pip install https://tvm-repo.s3-us-west-2.amazonaws.com/tvm-0.7.dev1-cp37-cp37m-linux_x86_64.whl
Our code runs on tvm-0.7-dev1 for now.
Finally, install MXNet’s CUDA version if GPUs are available [Chen et al., 2015]. Assume you are have CUDA 10.1 installed, then
pip install mxnet-cu101
You can change the 101 to match your CUDA version.
Once all packages are installed, you can open the Jupyter notebook by
jupyter notebook
At this point open http://localhost:8888 (which usually opens automatically) in the browser, then you can view and run the code in each section of the book.
1.1.3. Code¶
Throughout the book, we save reusable code blocks in the d2ltvm
package by adding the comment: “# Save to the d2ltvm package.”
before the code block. For example, the following code snippet shows the
libraries imported by d2ltvm.
# Save to the d2ltvm package.
import tvm
from tvm import te
import time
import timeit
import numpy as np
from matplotlib import pyplot as plt
from IPython import display
try:
import mxnet as mx
except:
pass