search

apple / tensorflow_macos

TensorFlow for macOS 11.0+ accelerated using Apple's ML Compute framework.
Other
3.66k stars 308 forks source link

Apple mac M1 Tensorflow 2 - Segmentation fault #251

Open ghost opened 3 years ago

ghost commented 3 years ago

I have Mac M1 chip machine. Installed M1 tensorflow setup : tensorflow_macos-0.1alpha3.tar.gz (https://github.com/apple/tensorflow_macos/releases) Trying to do custom object detection using 'ssd_mobilenet_v2_320x320_coco17_tpu' When I run 'model_main_tf2.py', the training stops immediately with following errors

2021-05-04 11:49:31.460883: I tensorflow/compiler/mlir/mlir_graph_optimization_pass.cc:116] None of the MLIR optimization passes are enabled (registered 2) 2021-05-04 11:49:31.463530: W tensorflow/core/platform/profile_utils/cpu_utils.cc:126] Failed to get CPU frequency: 0 Hz Fatal Python error: Segmentation fault

ShwetaTiwariRepo commented 3 years ago

i see similar error with resnet model architecture run on mnist dataset.

Niclas70 commented 3 years ago

Also struggling getting TensorFlow up running in M1. Tried several ways in Anaconda. Still not succeeded

IAINATTWATER commented 3 years ago

@Niclas70 make sure you're using Miniconda per the instructions. I assume you're also setting the env and activating it. I set it up following the instructions and able to run CNN/NN in Pycharm (either script or Jupyter). Also use the mlcompute directive to force gpu and is about 30% faster per step on the mnist fashion CNN.

Niclas70 commented 3 years ago

@Niclas70 make sure you're using Miniconda per the instructions. I assume you're also setting the env and activating it. I set it up following the instructions and able to run CNN/NN in Pycharm (either script or Jupyter). Also use the mlcompute directive to force gpu and is about 30% faster per step on the mnist fashion CNN.

Thanks, but got it to work. Not sure exactly what made it work but ensured right Python ver was used. The https://medium.com/gft-engineering/macbook-m1-tensorflow-on-jupyter-notebooks-6171e1f48060# was very good