areusch commented on a change in pull request #5655: URL: https://github.com/apache/incubator-tvm/pull/5655#discussion_r438270610
########## File path: tutorials/micro/micro_tflite.py ########## @@ -0,0 +1,219 @@ +# 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. +""" +.. _tutorial-micro-tflite: + +Micro TVM with TFLite Models +============================ +**Author**: `Tom Gall <https://github.com/tom-gall>`_ + +This tutorial is an introduction to working with MicroTVM and TFLite models with Relay. +""" +###################################################################### +# Setup +# ----- +# +# To get started, TFLite package needs to be installed as prerequisite. +# +# install tflite +# .. code-block:: bash +# +# pip install tflite=2.1.0 --user +# +# or you could generate TFLite package yourself. The steps are the following: +# +# Get the flatc compiler. +# Please refer to https://github.com/google/flatbuffers for details +# and make sure it is properly installed. +# +# .. code-block:: bash +# +# flatc --version +# +# Get the TFLite schema. +# +# .. code-block:: bash +# +# wget https://raw.githubusercontent.com/tensorflow/tensorflow/r1.13/tensorflow/lite/schema/schema.fbs +# +# Generate TFLite package. +# +# .. code-block:: bash +# +# flatc --python schema.fbs +# +# Add current folder (which contains generated tflite module) to PYTHONPATH. +# +# .. code-block:: bash +# +# export PYTHONPATH=${PYTHONPATH:+$PYTHONPATH:}$(pwd) +# +# To validate that the TFLite package was installed successfully, ``python -c "import tflite"`` +# +# CMSIS needs to be downloaded and the CMSIS_ST_PATH environment variable setup +# This tutorial only supports the STM32F7xx series of boards. +# Download from : https://www.st.com/en/embedded-software/stm32cubef7.html +# After you've expanded the zip file +# +# .. code-block:: bash +# +# export CMSIS_ST_PATH=/path/to/STM32Cube_FW_F7_V1.16.0/Drivers/CMSIS +# +# Next we need to download a pretrained TFLite model. When working with microcontrollers Review comment: it might read better if we add a header "Download Pre-Trained TFLite Model" before this line. ########## File path: tutorials/micro/micro_tflite.py ########## @@ -0,0 +1,219 @@ +# 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. +""" +.. _tutorial-micro-tflite: + +Micro TVM with TFLite Models +============================ +**Author**: `Tom Gall <https://github.com/tom-gall>`_ + +This tutorial is an introduction to working with MicroTVM and TFLite models with Relay. +""" +###################################################################### +# Setup +# ----- +# +# To get started, TFLite package needs to be installed as prerequisite. +# +# install tflite +# .. code-block:: bash +# +# pip install tflite=2.1.0 --user +# +# or you could generate TFLite package yourself. The steps are the following: +# +# Get the flatc compiler. +# Please refer to https://github.com/google/flatbuffers for details +# and make sure it is properly installed. +# +# .. code-block:: bash +# +# flatc --version +# +# Get the TFLite schema. +# +# .. code-block:: bash +# +# wget https://raw.githubusercontent.com/tensorflow/tensorflow/r1.13/tensorflow/lite/schema/schema.fbs +# +# Generate TFLite package. +# +# .. code-block:: bash +# +# flatc --python schema.fbs +# +# Add current folder (which contains generated tflite module) to PYTHONPATH. +# +# .. code-block:: bash +# +# export PYTHONPATH=${PYTHONPATH:+$PYTHONPATH:}$(pwd) +# +# To validate that the TFLite package was installed successfully, ``python -c "import tflite"`` +# +# CMSIS needs to be downloaded and the CMSIS_ST_PATH environment variable setup +# This tutorial only supports the STM32F7xx series of boards. +# Download from : https://www.st.com/en/embedded-software/stm32cubef7.html +# After you've expanded the zip file +# +# .. code-block:: bash +# +# export CMSIS_ST_PATH=/path/to/STM32Cube_FW_F7_V1.16.0/Drivers/CMSIS +# +# Next we need to download a pretrained TFLite model. When working with microcontrollers +# you need to be mindful these are highly resource constrained devices as such standard +# models like Mobilenet may not fit into their modest memory. +# +# For this tutorial, we'll make use of one of the TF Micro example models. +# +# If you wish to replicate the training steps see: +# https://github.com/tensorflow/tensorflow/tree/master/tensorflow/lite/micro/examples/hello_world/train +# +# .. code-block:: bash +# +# if you download the example pretrained model from +# wget https://storage.googleapis.com/download.tensorflow.org/models/tflite/micro/hello_world_2020_04_13.zip +# unzip hello_world_2020_04_13.zip +# this will fail due to an unimplemented opcode (114) +# I've saved an older version of the pre-trailed model and made it available on linaro.org + +###################################################################### +# Python imports for tvm, numpy etc Review comment: This text renders as a header--I think you'll need to indent here to mark this text as a large code block. Here, I think it would make sense to include another header akin to "Import the TFLite Model." ########## File path: tutorials/micro/micro_tflite.py ########## @@ -0,0 +1,219 @@ +# 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. +""" +.. _tutorial-micro-tflite: + +Micro TVM with TFLite Models +============================ +**Author**: `Tom Gall <https://github.com/tom-gall>`_ + +This tutorial is an introduction to working with MicroTVM and TFLite models with Relay. +""" +###################################################################### +# Setup +# ----- +# +# To get started, TFLite package needs to be installed as prerequisite. +# +# install tflite +# .. code-block:: bash +# +# pip install tflite=2.1.0 --user +# +# or you could generate TFLite package yourself. The steps are the following: +# +# Get the flatc compiler. +# Please refer to https://github.com/google/flatbuffers for details +# and make sure it is properly installed. +# +# .. code-block:: bash +# +# flatc --version +# +# Get the TFLite schema. +# +# .. code-block:: bash +# +# wget https://raw.githubusercontent.com/tensorflow/tensorflow/r1.13/tensorflow/lite/schema/schema.fbs +# +# Generate TFLite package. +# +# .. code-block:: bash +# +# flatc --python schema.fbs +# +# Add current folder (which contains generated tflite module) to PYTHONPATH. +# +# .. code-block:: bash +# +# export PYTHONPATH=${PYTHONPATH:+$PYTHONPATH:}$(pwd) +# +# To validate that the TFLite package was installed successfully, ``python -c "import tflite"`` +# +# CMSIS needs to be downloaded and the CMSIS_ST_PATH environment variable setup +# This tutorial only supports the STM32F7xx series of boards. +# Download from : https://www.st.com/en/embedded-software/stm32cubef7.html +# After you've expanded the zip file +# +# .. code-block:: bash +# +# export CMSIS_ST_PATH=/path/to/STM32Cube_FW_F7_V1.16.0/Drivers/CMSIS +# +# Next we need to download a pretrained TFLite model. When working with microcontrollers +# you need to be mindful these are highly resource constrained devices as such standard +# models like Mobilenet may not fit into their modest memory. +# +# For this tutorial, we'll make use of one of the TF Micro example models. +# +# If you wish to replicate the training steps see: +# https://github.com/tensorflow/tensorflow/tree/master/tensorflow/lite/micro/examples/hello_world/train +# +# .. code-block:: bash +# +# if you download the example pretrained model from +# wget https://storage.googleapis.com/download.tensorflow.org/models/tflite/micro/hello_world_2020_04_13.zip +# unzip hello_world_2020_04_13.zip +# this will fail due to an unimplemented opcode (114) +# I've saved an older version of the pre-trailed model and made it available on linaro.org + +###################################################################### +# Python imports for tvm, numpy etc +# ---------------------------------------------- +import os +import numpy as np +import tvm +import tvm.micro as micro +import requests + +from tvm.contrib import graph_runtime, util +from tvm import relay + + +###################################################################### +# Load the pretrained TFLite model from a file in your current +# directory into a buffer +model_url = 'https://people.linaro.org/~tom.gall/sine_model.tflite' Review comment: During the CI, all of this code will get executed. @tqchen i'm not sure we'd like to include external dependencies in the CI. Is it okay to place this ~2.5kB file in the tutorials dir? ########## File path: tutorials/micro/micro_tflite.py ########## @@ -0,0 +1,219 @@ +# 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. +""" +.. _tutorial-micro-tflite: + +Micro TVM with TFLite Models +============================ +**Author**: `Tom Gall <https://github.com/tom-gall>`_ + +This tutorial is an introduction to working with MicroTVM and TFLite models with Relay. +""" +###################################################################### +# Setup +# ----- +# +# To get started, TFLite package needs to be installed as prerequisite. +# +# install tflite +# .. code-block:: bash +# +# pip install tflite=2.1.0 --user +# +# or you could generate TFLite package yourself. The steps are the following: +# +# Get the flatc compiler. +# Please refer to https://github.com/google/flatbuffers for details +# and make sure it is properly installed. +# +# .. code-block:: bash +# +# flatc --version +# +# Get the TFLite schema. +# +# .. code-block:: bash +# +# wget https://raw.githubusercontent.com/tensorflow/tensorflow/r1.13/tensorflow/lite/schema/schema.fbs +# +# Generate TFLite package. +# +# .. code-block:: bash +# +# flatc --python schema.fbs +# +# Add current folder (which contains generated tflite module) to PYTHONPATH. +# +# .. code-block:: bash +# +# export PYTHONPATH=${PYTHONPATH:+$PYTHONPATH:}$(pwd) +# +# To validate that the TFLite package was installed successfully, ``python -c "import tflite"`` +# +# CMSIS needs to be downloaded and the CMSIS_ST_PATH environment variable setup +# This tutorial only supports the STM32F7xx series of boards. +# Download from : https://www.st.com/en/embedded-software/stm32cubef7.html +# After you've expanded the zip file +# +# .. code-block:: bash +# +# export CMSIS_ST_PATH=/path/to/STM32Cube_FW_F7_V1.16.0/Drivers/CMSIS +# +# Next we need to download a pretrained TFLite model. When working with microcontrollers +# you need to be mindful these are highly resource constrained devices as such standard +# models like Mobilenet may not fit into their modest memory. +# +# For this tutorial, we'll make use of one of the TF Micro example models. +# +# If you wish to replicate the training steps see: +# https://github.com/tensorflow/tensorflow/tree/master/tensorflow/lite/micro/examples/hello_world/train +# +# .. code-block:: bash +# +# if you download the example pretrained model from +# wget https://storage.googleapis.com/download.tensorflow.org/models/tflite/micro/hello_world_2020_04_13.zip +# unzip hello_world_2020_04_13.zip +# this will fail due to an unimplemented opcode (114) +# I've saved an older version of the pre-trailed model and made it available on linaro.org + +###################################################################### +# Python imports for tvm, numpy etc +# ---------------------------------------------- +import os +import numpy as np +import tvm +import tvm.micro as micro +import requests + +from tvm.contrib import graph_runtime, util +from tvm import relay + + +###################################################################### +# Load the pretrained TFLite model from a file in your current +# directory into a buffer +model_url = 'https://people.linaro.org/~tom.gall/sine_model.tflite' +model_file = 'sine_model.tflite' +r = requests.get(model_url, allow_redirects=True) + +open(model_file,'wb').write(r.content) +###################################################################### +# Uncomment the following code to load the model from a local +# directory +# Load the pretrained TFLite model from a file in your current +# directory into a buffer +# model_dir ="./" +# tflite_model_file = os.path.join(model_dir, "sine_model.tflite") +tflite_model_buf = open(model_file, "rb").read() + +###################################################################### +# Using the buffer, transform into a tflite model python object +try: + import tflite + tflite_model = tflite.Model.GetRootAsModel(tflite_model_buf, 0) +except AttributeError: + import tflite.Model + tflite_model = tflite.Model.Model.GetRootAsModel(tflite_model_buf, 0) + +###################################################################### +# Print out the version of the model +version = tflite_model.Version() +print ("Model Version: " + str(version)) + + +###################################################################### +# Setup the device config which is what will be used to communicate +# with the microcontroller (a STM32F746 Discovery board) +TARGET = 'c -device=micro_dev' +dev_config = micro.device.arm.stm32f746xx.generate_config("127.0.0.1", 6666) + + +###################################################################### +# Parse the python model object to convert it into a relay module +# and weights +# It is important to note that the input tensor name must match what +# is contained in the model. +# If you are unsure what that might be, this can be discovered by using +# the visualize.py script within the Tensorflow project. +# See : How do I inspect a .tflite file? https://www.tensorflow.org/lite/guide/faq +input_tensor = "dense_4_input" +input_shape = (1,) +input_dtype = "float32" + +mod, params = relay.frontend.from_tflite(tflite_model, + shape_dict={input_tensor: input_shape}, + dtype_dict={input_tensor: input_dtype}) + +###################################################################### Review comment: here you could add another section "Run the imported model on device" ---------------------------------------------------------------- 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: us...@infra.apache.org
