[GitHub] [incubator-tvm] leandron commented on a change in pull request #6597: [tvmc][docs] Getting started tutorial for TVMC

[GitBox <gi...@apache.org>]

leandron commented on a change in pull request #6597:
URL: https://github.com/apache/incubator-tvm/pull/6597#discussion_r498056155



##########
File path: tutorials/get_started/tvmc_command_line_driver.py
##########
@@ -0,0 +1,313 @@
+# Licensed to the Apache Software Foundation (ASF) under one
+# or more contributor license agreements.  See the NOTICE file
+# distributed with this work for additional information
+# regarding copyright ownership.  The ASF licenses this file
+# to you under the Apache License, Version 2.0 (the
+# "License"); you may not use this file except in compliance
+# with the License.  You may obtain a copy of the License at
+#
+#   http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing,
+# software distributed under the License is distributed on an
+# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+# KIND, either express or implied.  See the License for the
+# specific language governing permissions and limitations
+# under the License.
+"""
+Getting Started with TVM command line driver - TVMC
+===================================================
+**Authors**:
+`Leandro Nunes <https://github.com/leandron>`_,
+`Matthew Barrett <https://github.com/mbaret>`_
+
+This tutorial is an introduction to working with TVMC, the TVM command
+line driver.
+
+In this tutorial we are going to use TVMC to perform common tasks with
+TVM via a command line interface.
+"""
+
+######################################################################
+# Using TVMC
+# ----------
+#
+# TVMC is a Python application, part of the TVM Python package.
+# When you install TVM using a Python package, you will get TVMC as
+# command line application ``tvmc``.
+#
+# Alternatively, if you have TVM as a Python module on your
+# ``$PYTHONPATH``,you can access the command line driver functionality
+# via the executable python module, ``python -m tvm.driver.tvmc``.
+#
+# For simplicity, this tutorial will mention TVMC command line using
+# ``tvmc <options>``, but the same results can be obtained with
+# ``python -m tvm.driver.tvmc <options>``.
+#
+# You can check the help page using:
+#
+# .. code-block:: bash
+#
+#   tvmc --help
+#
+#
+# As you can see in the help page, the main features are
+# accessible via the subcommands ``tune``, ``compile`` and ``run``.
+# To read about specific options under a given subcommand, use
+# ``tvmc <subcommand> --help``.
+#
+# By default, TVMC will only display error messages. In case you want
+# to increase verbosity (more messages), you can do that with:
+#
+# .. code-block:: bash
+#
+#   tvmc -vvv <subcommand>
+#
+#
+# In the following sections we will use TVMC to tune, compile and
+# run a model. But first, we need a model.
+#
+
+
+######################################################################
+# Obtaining the model
+# -------------------
+#
+# We are going to use ResNet-50 V2 as an example to experiment with TVMC.
+# The version below is in ONNX format. To download the file, you can use
+# the command below:
+#
+# .. code-block:: bash
+#
+#   wget https://github.com/onnx/models/raw/master/vision/classification/resnet/model/resnet50-v2-7.onnx
+#
+#
+
+######################################################################
+# .. note:: Supported model formats
+#
+#   TVMC supports models created with Keras, ONNX, TensorFlow, TFLite
+#   and Torch. Use the option``--model-format`` if you need to
+#   explicitly provide the model format you are using. See ``tvmc
+#   compile --help`` for more information.
+#
+
+######################################################################
+# Tuning the model
+# ----------------
+#
+# Once we've downloaded ResNet-50, it is time to run TVM's auto-tuner.
+# Tuning in TVM refers to the process by which a model is optimized
+# to run faster on a given target. This differs from training or
+# fine-tuning in that it does not affect the accuracy of the model,
+# only the runtime performance.
+#
+# As part of the tuning process, TVM will try running many different
+# operator implementation variants to see which perform best. The
+# results of these runs are stored in a tuning records file, which is
+# ultimately the output of the ``tune`` subcommand.
+#
+# In the simplest form, tuning requires you to provide three things:
+#
+# - the target specification of the device you intend to run this model on;
+# - the path to an output file in which the tuning records will be stored, and finally,
+# - a path to the model to be tuned.
+#
+#
+# The example below demonstrates how that works in practice:
+#
+# .. code-block:: bash
+#
+#   tvmc tune \
+#     --target "llvm" \
+#     --output autotuner_records.json \
+#     resnet50-v2-7.onnx
+#
+#
+# Tuning sessions can take a long time, so ``tvmc tune`` offers many options to
+# customize your tuning process, in terms of number of repetitions (``--repeat`` and
+# ``--number``, for example), the tuning algorithm to be use, and so on.
+# Check ``tvmc tune --help`` for more information.
+#
+# As an output of the tuning process above, we obtained the tuning records stored
+# in ``autotuner_records.json``. This file can be used in two ways: as an input
+# to further tuning (via ``tvmc tune --tuning-records``), or as an input to the
+# compiler. The compiler will use the results to generate high performance code
+# for the model on your specified target.
+#
+# So now let's see how we can invoke the compiler using TVMC.
+#
+
+######################################################################
+# Compiling the model
+# -------------------
+#
+# To compile our model using TVM, we use ``tvmc compile``. The output we get from the
+# compilation process is a TAR package that can be used to run our model on the
+# target device.
+#
+# As previously mentioned, we can use tuning records in order to help the compilation
+# process to generate code that performs better for a given platform. To do that
+# we can use ``--tuning-records``. Check ``tvmc compile --help`` for more options.
+#
+# .. code-block:: bash
+#
+#   tvmc compile \
+#     --target "llvm" \
+#     --output compiled_module.tar \
+#     --tuning-records autotuner_records.json \
+#     resnet50-v2-7.onnx
+#
+# Once compilation finishes, the output ``compiled_module.tar`` will be created. This
+# can be directly loaded by your application and run via the TVM runtime APIs.
+# Alternatively, you can make use of the ``tvmc run`` command to quickly generate
+# predictions using the compiled module without having to write such an application.
+#
+
+
+######################################################################
+# Input pre-processing
+# --------------------
+#
+# In order to generate predictions, we will need two things:
+#
+# - the compiled module, which we just produced;
+# - a valid input to the model
+#
+# Each model is particular when it comes to expected tensor shapes, formats and data
+# types. For this reason, most models require some pre and
+# post processing, to ensure the input(s) is valid and to interpret the output(s).
+#
+# In TVMC, we adopted NumPy's ``.npz`` format for both input and output data.
+# This is a well-supported NumPy format to serialize multiple arrays into a file.

Review comment:
       I agree with your point. Yes, there are plans to have such features in next versions. For ImageNet it it looks trivial enough, the challenge is that is should be flexible to allow users to bring their own pre/post processing. We'll certainly discuss more about that, so I'll take no action for now.




----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

For queries about this service, please contact Infrastructure at:
users@infra.apache.org