Skip to content

Latest commit

 

History

History
494 lines (370 loc) · 17.2 KB

tensorflow.md

File metadata and controls

494 lines (370 loc) · 17.2 KB
Raw

TensorFlow on Apple M3 Pro

https://www.tensorflow.org/install/source#macos https://www.tensorflow.org/install/pip

Guidance:

Version Python Compiler Build tools
tensorflow-2.16.1 3.9-3.12 Clang 17.0.6 (xcode 13.6) Bazel 6.5.0
tensorflow-2.15.0 3.9-3.11 Clang 16.0.0 (xcode 10.15) Bazel 6.1.0

PROBLEM TO SOLVE

similar to tensorflow/text#823

ERROR: Could not find a version that satisfies the requirement tensorflow_text (from versions: none)
ERROR: No matching distribution found for tensorflow_text

Xcode v15.1

% /usr/bin/xcodebuild -version
xcode-select: error: tool 'xcodebuild' requires Xcode, but active developer directory '/Library/Developer/CommandLineTools' is a command line tools instance

% pkgutil --pkg-info=com.apple.pkg.CLTools_Executables | grep version
version: 15.1.0.0.1.1700200546

Download Xcode v15 from App Store

 % /usr/bin/xcodebuild -version                                       
Xcode 15.1
Build version 15C65

% xcrun -f ld                                                        
/Applications/Xcode.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/ld

% ld -v
@(#)PROGRAM:ld  PROJECT:dyld-1022.1
BUILD 13:21:42 Nov 10 2023
configured to support archs: armv6 armv7 armv7s arm64 arm64e arm64_32 i386 x86_64 x86_64h
will use ld-classic for: armv6 armv7 armv7s arm64_32 i386 armv6m armv7k armv7m armv7em
LTO support using: LLVM version 15.0.0 (static support for 29, runtime is 29)
TAPI support using: Apple TAPI version 15.0.0 (tapi-1500.0.12.8)
Library search paths:
Framework search paths:

^ LLVM version 15.0.0

Clang v16.0

https://www.tensorflow.org/install/source#install_clang_recommended_linux_only

Clang is a C/C++/Objective-C compiler that is compiled in C++ based on LLVM. It is the default compiler to build TensorFlow starting with TensorFlow 2.13. The current supported version is LLVM/Clang 16.

% /usr/bin/clang -v
Apple clang version 15.0.0 (clang-1500.1.0.2.5)
Target: arm64-apple-darwin23.2.0
Thread model: posix
InstalledDir: /Applications/Xcode.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin

Macports https://www.macports.org/

~ % source ~/.zprofile # Macports already added PATH
~ % sudo port install clang-16
~ % port contents clang-16

~ % /opt/local/bin/clang-mp-16 -v
clang version 16.0.6
Target: arm64-apple-darwin23.2.0
Thread model: posix
InstalledDir: /opt/local/libexec/llvm-16/bin
~ % echo "" >> ~/.zprofile
~ % echo "# Clang (v16.0) for Tensorflow v2.15" >> ~/.zprofile
~ % echo "PATH=\"/opt/local/libexec/llvm-16/bin:\${PATH}\"" >> ~/.zprofile
~ % echo "export PATH" >> ~/.zprofile
~ % source ~/.zprofile

and now

~ % clang -v
clang version 16.0.6
Target: arm64-apple-darwin23.2.0
Thread model: posix
InstalledDir: /opt/local/libexec/llvm-16/bin

Bazel v6.1.0

The project you're trying to build requires Bazel 6.1.0.

Make sure to install the correct Bazel version from TensorFlow's **.bazelversion file.

https://bazel.build/install/bazelisk#install-with-installer-mac-os-x

~ % export BAZEL_VERSION=6.1.0
~ % curl -fLO "https://github.com/bazelbuild/bazel/releases/download/$BAZEL_VERSION/bazel-$BAZEL_VERSION-installer-darwin-arm64.sh"
~ % chmod +x "bazel-$BAZEL_VERSION-installer-darwin-arm64.sh"
~ % ./bazel-$BAZEL_VERSION-installer-darwin-arm64.sh --user
...
Bazel is now installed!

Make sure you have "/Users/marksusol/bin" in your path.

Let's add the Bazel PATH to our ~/.zprofile file.

~ % echo "" >> ~/.zprofile
~ % echo "# Bazel (user)" >> ~/.zprofile
~ % echo "PATH=\"/Users/marksusol/bin:\${PATH}\"" >> ~/.zprofile
~ % echo "export PATH" >> ~/.zprofile
~ % echo "export MACOSX_DEPLOYMENT_TARGET=14.2 " >> ~/.zprofile
~ % source ~/.zprofile

You can confirm Bazel is installed successfully by running the following command:

~ % bazel --version
bazel 6.1.0

Building TensorFlow

~ % pip --version
pip 23.3.2 from /Library/Frameworks/Python.framework/Versions/3.11/lib/python3.11/site-packages/pip (python 3.11)

TensorFlow v2.15.0

https://github.com/tensorflow/tensorflow/tree/r2.15

To set TensorFlow's Python version, add to ~/.zprofile:

~ % echo "" >> ~/.zprofile
~ % echo "# Setting Tensorflow python version" >> ~/.zprofile
~ % echo "export TF_PYTHON_VERSION=3.11" >> ~/.zprofile
~ % source ~/.zprofile

Now let's clone the TensorFlow repository into our local directory used for git clones.

~ % cd /DataScience/GitHub
GitHub % git clone --branch v2.15.0 https://github.com/tensorflow/tensorflow.git
GitHub % cd tensorflow
tensorflow % which python3        
/Library/Frameworks/Python.framework/Versions/3.11/bin/python3

tensorflow % python3 -m venv .venv
tensorflow % source .venv/bin/activate
(.venv) tensorflow % pip install --upgrade pip
(.venv) tensorflow % pip3 install numpy==1.26.2 wheel==0.42.0 packaging==23.2 requests==2.31.0 opt-einsum==3.3.0 keras==2.15
(.venv) tensorflow % pip3 install Keras-Preprocessing==1.1.2 --no-deps
(.venv) tensorflow % ./configure

You have bazel 6.1.0 installed.
Please specify the location of python. [Default is /Users/marksusol/DataScience/GitHub/tensorflow/.venv/bin/python3]: /Users/marksusol/DataScience/GitHub/tensorflow/.venv/bin/python3 


Found possible Python library paths:
  /Users/marksusol/DataScience/GitHub/tensorflow/.venv/lib/python3.11/site-packages
Please input the desired Python library path to use.  Default is [/Users/marksusol/DataScience/GitHub/tensorflow/.venv/lib/python3.11/site-packages]
/Users/marksusol/DataScience/GitHub/tensorflow/.venv/lib/python3.11/site-packages
Do you wish to build TensorFlow with ROCm support? [y/N]: N
No ROCm support will be enabled for TensorFlow.

Do you wish to build TensorFlow with CUDA support? [y/N]: N
No CUDA support will be enabled for TensorFlow.

Please specify optimization flags to use during compilation when bazel option "--config=opt" is specified [Default is -Wno-sign-compare]: -Wno-sign-compare

Would you like to interactively configure ./WORKSPACE for Android builds? [y/N]: N
Not configuring the WORKSPACE for Android builds.

Do you wish to build TensorFlow with iOS support? [y/N]: N
No iOS support will be enabled for TensorFlow.

Preconfigured Bazel build configs. You can use any of the below by adding "--config=<>" to your build command. See .bazelrc for more details.
        --config=mkl            # Build with MKL support.
        --config=mkl_aarch64    # Build with oneDNN and Compute Library for the Arm Architecture (ACL).
        --config=monolithic     # Config for mostly static monolithic build.
        --config=numa           # Build with NUMA support.
        --config=dynamic_kernels        # (Experimental) Build kernels into separate shared objects.
        --config=v1             # Build with TensorFlow 1 API instead of TF 2 API.
Preconfigured Bazel build configs to DISABLE default on features:
        --config=nogcp          # Disable GCP support.
        --config=nonccl         # Disable NVIDIA NCCL support.
Configuration finished

Now lets 'bazel build' ..

(.venv) tensorflow % bazel build -c opt --config=macos_arm64 \
  --macos_minimum_os="${MACOSX_DEPLOYMENT_TARGET}" //tensorflow/tools/pip_package:build_pip_package

...
INFO: Found applicable config definition build:macos_arm64 in file /Users/marksusol/DataScience/GitHub/tensorflow/.bazelrc: --cpu=darwin_arm64 --macos_minimum_os=11.0
...
Target //tensorflow/tools/pip_package:build_pip_package up-to-date:
  bazel-bin/tensorflow/tools/pip_package/build_pip_package
INFO: Elapsed time: 5059.060s, Critical Path: 538.10s
INFO: 22267 processes: 5077 internal, 17190 local.
INFO: Build completed successfully, 22267 total actions

(.venv) tensorflow % ./bazel-bin/tensorflow/tools/pip_package/build_pip_package /tmp/tensorflow_pkg

(.venv) tensorflow % ls /tmp/tensorflow_pkg
tensorflow-2.15.0-cp311-cp311-macosx_14_0_arm64.whl

(.venv) tensorflow % deactivate

TensorFlow I/O v0.34.0

GitHub % git clone --branch v0.34.0 https://github.com/tensorflow/io.git
GitHub % cd io
io % python3 -m venv .venv
io % source .venv/bin/activate
(.venv) io % pip install --upgrade pip
(.venv) io % pip3 install setuptools==65.5.0 wheel==0.42.0
(.venv) io % python3 setup.py -q bdist_wheel --project tensorflow_io_gcs_filesystem
(.venv) io % ls -l dist                                                                        

-rw-r--r--@ 1 marksusol  staff  12908 Jan  9 09:51 tensorflow_io_gcs_filesystem-0.34.0-cp311-cp311-macosx_14_0_universal2.whl

(.venv) io % deactivate

TensorFlow Text v2.15.0

Copy the tensorflow and tensorflow-io wheels into this directory. Install tensorflow-io first, then tensorflow.

GitHub % git clone --branch v2.15.0 https://github.com/tensorflow/text.git
GitHub % cd text
text % python3 -m venv .venv
text % source .venv/bin/activate
(.venv) text % pip install --upgrade pip
(.venv) text % pip3 install setuptools==65.5.0 wheel==0.42.0

(.venv) text % cp ../io/dist/tensorflow_io_gcs_filesystem-0.34.0-cp311-cp311-macosx_14_0_universal2.whl \
  tensorflow_io_gcs_filesystem-0.34.0-cp311-cp311-macosx_14_0_universal2.whl
(.venv) text % cp /tmp/tensorflow_pkg/tensorflow-2.15.0-cp311-cp311-macosx_14_0_arm64.whl \
  tensorflow-2.15.0-cp311-cp311-macosx_14_0_arm64.whl
  
(.venv) text % pip3 install tensorflow_io_gcs_filesystem-0.34.0-cp311-cp311-macosx_14_0_universal2.whl
(.venv) text % pip3 install tensorflow-2.15.0-cp311-cp311-macosx_14_0_arm64.whl

(.venv) text % pip3 list | grep tensorflow

tensorflow                   2.15.0
tensorflow-estimator         2.15.0
tensorflow-io-gcs-filesystem 0.34.0

Next you need to edit ./oss_scripts/run_build.sh and add the bazel flags:

doesn't work?

# Build the pip package.
bazel build -c opt --config=macos_arm64 --enable_runfiles \
  --macos_minimum_os="${MACOSX_DEPLOYMENT_TARGET}" oss_scripts/pip_package:build_pip_package

and now ...

(.venv) text % ./oss_scripts/run_build.sh
(.venv) text % pip3 install tensorflow_text-2.15.0-cp311-cp311-macosx_11_0_arm64.whl
(.venv) text % pip list | grep tensorflow

tensorflow                   2.15.0
tensorflow-estimator         2.15.0
tensorflow-hub               0.15.0
tensorflow-io-gcs-filesystem 0.34.0
tensorflow-macos             2.15.0
tensorflow-text              2.15.0

copy over for saving in repository

(.venv) text % cp tensorflow_io_gcs_filesystem-0.34.0-cp311-cp311-macosx_14_0_universal2.whl \
  ../msusol/jupyter-notebook-on-macos/dist/tensorflow_io_gcs_filesystem-0.34.0-cp311-cp311-macosx_14_0_universal2.whl
(.venv) text % cp tensorflow-2.15.0-cp311-cp311-macosx_14_0_arm64.whl \
  ../msusol/jupyter-notebook-on-macos/dist/tensorflow-2.15.0-cp311-cp311-macosx_14_0_arm64.whl
(.venv) text % cp tensorflow_text-2.15.0-cp311-cp311-macosx_11_0_arm64.whl \
  ../msusol/jupyter-notebook-on-macos/dist/tensorflow_text-2.15.0-cp311-cp311-macosx_11_0_arm64.whl

Now you can copy all three wheels into a new directory and install them and watch the magic.

dist % pip3 install tensorflow_io_gcs_filesystem-0.35.0-cp311-cp311-macosx_14_0_universal2.whl
dist % pip3 install tensorflow-2.15.0-cp311-cp311-macosx_14_0_arm64.whl
dist % pip3 install tensorflow_text-2.15.0-cp311-cp311-macosx_11_0_arm64.whl

tensorflow-text

I also want to give credit to TeoZosa for his repo here. This didn’t quite work for me, but it gave me some ideas and helped me figure some things out.

TensorFlow Metal & GPU Support

~ % python3 -m venv .venv-metal
~ % source .venv-metal/bin/activate
(.venv-metal) % pip install --upgrade pip

The h5py package depends on hdf5 which cannot be installed through pip.

handle the h5py package and its non-Python dependency, hdf5

Since hdf5 cannot be installed through pip,

HDF5 for Python is all you need to get TensorFlow running with GPU support on your M3 Mac is to install hdf5 through MacPorts, and then install both tensorflow-macos (done above) and tensorflow-metal through pip.

See also:

Check state before installing hdf5:

import tensorflow as tf

for device in ['CPU', 'GPU']:
    print(tf.config.list_physical_devices(device))

[PhysicalDevice(name='/physical_device:CPU:0', device_type='CPU')]
[]

and now let's install hdf5

~ % sudo port install hdf5

--->  Computing dependencies for hdf5
The following dependencies will be installed: 
 libaec
 libgcc
 libgcc13
...
  hdf5 has the following notes:
    Mac users may need to set the environment variable "HDF5_USE_FILE_LOCKING" to the five-character string "FALSE" when accessing network mounted files. This is an application run-time setting, not a
    configure or build setting. Otherwise errors such as "unable to open file" or "HDF5 error" may be encountered.

https://github.com/tensorflow/tensorflow/blob/r2.15/tensorflow/tools/pip_package/setup.py#L90,93,109

REQUIRED_PACKAGES = [
    ...
    'h5py >= 2.9.0',
     ...
    'numpy >= 1.23.5, < 2.0.0',
    ...
    'grpcio >= 1.24.3, < 2.0'
    ...
]
~ % pip list | grep -E 'grpcio|h5py|numpy'

grpcio                        1.60.0
h5py                          3.10.0
numpy                         1.26.2

# ~ % pip3 install -U "grpcio>=1.59.0,<2.0" "h5py>=3.10.0,<3.11" "numpy>=1.23.5,<2.0.0"

Now let's install Tensorflow from our dist files in this venv:

(.venv-metal) % pip install --upgrade pip
(.venv-metal) % cd dist
(.venv-metal) dist % pip3 install tensorflow_io_gcs_filesystem-0.35.0-cp311-cp311-macosx_14_0_universal2.whl
(.venv-metal) dist % pip3 install tensorflow-2.15.0-cp311-cp311-macosx_14_0_arm64.whl
(.venv-metal) dist % pip3 install tensorflow_text-2.15.0-cp311-cp311-macosx_11_0_arm64.whl

Launch your colab instance now from this .venv-metal environment and run the following.

GPU

import tensorflow as tf

# Ensure we see the GPU in device list.
print('Visible Devices: ', tf.config.get_visible_devices())

cifar = tf.keras.datasets.cifar100
(x_train, y_train), (x_test, y_test) = cifar.load_data()
model = tf.keras.applications.ResNet50(
  include_top=True,
  weights=None,
  input_shape=(32, 32, 3),
  classes=100,)

loss_fn = tf.keras.losses.SparseCategoricalCrossentropy(from_logits=False)
model.compile(optimizer="adam", loss=loss_fn, metrics=["accuracy"])

with tf.device('/device:GPU:0'):
  model.fit(x_train, y_train, epochs=1, batch_size=128)
Visible Devices:  [PhysicalDevice(name='/physical_device:CPU:0', device_type='CPU'), PhysicalDevice(name='/physical_device:GPU:0', device_type='GPU')]
2024-01-11 11:39:24.910238: I metal_plugin/src/device/metal_device.cc:1154] Metal device set to: Apple M3 Pro
2024-01-11 11:39:24.910283: I metal_plugin/src/device/metal_device.cc:296] systemMemory: 18.00 GB
2024-01-11 11:39:24.910292: I metal_plugin/src/device/metal_device.cc:313] maxCacheSize: 6.00 GB
2024-01-11 11:39:24.910580: I tensorflow/core/common_runtime/pluggable_device/pluggable_device_factory.cc:306] Could not identify NUMA node of platform GPU ID 0, defaulting to 0. Your kernel may not have been built with NUMA support.
2024-01-11 11:39:24.910803: I tensorflow/core/common_runtime/pluggable_device/pluggable_device_factory.cc:272] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 0 MB memory) -> physical PluggableDevice (device: 0, name: METAL, pci bus id: <undefined>)

391/391 [==============================] - 42s 96ms/step - loss: 4.5789 - accuracy: 0.0778

tensorflow-cpu-gpu-activity

GPU evaluation process available in a Jupyter notebook: TensorFlowMetal

CPU

Now let us specifically remove the GPU from visible devices to test CPU only.

import tensorflow as tf
# Removes GPU from list, i.e. []
tf.config.set_visible_devices([], 'GPU')
print('Visible Devices: ', tf.config.get_visible_devices())

cifar = tf.keras.datasets.cifar100
(x_train, y_train), (x_test, y_test) = cifar.load_data()
model = tf.keras.applications.ResNet50(
  include_top=True,
  weights=None,
  input_shape=(32, 32, 3),
  classes=100,)

loss_fn = tf.keras.losses.SparseCategoricalCrossentropy(from_logits=False)
model.compile(optimizer="adam", loss=loss_fn, metrics=["accuracy"])

with tf.device('/device:CPU:0'):
  model.fit(x_train, y_train, epochs=1, batch_size=128)
Visible Devices:  [PhysicalDevice(name='/physical_device:CPU:0', device_type='CPU')]
391/391 [==============================] - 393s 1s/step - loss: 4.6609 - accuracy: 0.0720
CPU times: user 17min 44s, sys: 2min 35s, total: 20min 19s
Wall time: 6min 34s

tensorflow-cpu-only-activity