Closed Danliran closed 4 years ago
I alse post this issue in tensorflow https://github.com/tensorflow/tensorflow/issues/40213
Unfortunately, tf_cnn_benchmarks is no longer maintained so its unlikely it will get resolved. I recommend trying the TensorFlow Official Models instead. I'm unsure what the state of aarch64 in TensorFlow however, so am unsure if the Official Models will work either.
@gunan do you know how well supported aarch64 is in TensorFlow?
I wouldn't know. @petewarden may?
System information
Have I written custom code (as opposed to using a stock example script provided in TensorFlow):no OS Platform and Distribution (e.g., Linux Ubuntu 16.04):ubuntu 18.04 Mobile device (e.g. iPhone 8, Pixel 2, Samsung Galaxy) if the issue happens on mobile device: no TensorFlow installed from (source or binary): source TensorFlow version (use command below):1.15.0 Python version:2.7 Bazel version (if compiling from source):0.26.0 GCC/Compiler version (if compiling from source):7.4.0 CUDA/cuDNN version: no GPU model and memory:no Describe the current behavior
I used tensorflow/benchmark to train resnet50 by tf_cnn_benchmarks.py on my aarch64 platform. command line: python tf_cnn_benchmarks.py --device=cpu --data_format=NHWC --optimizer=sgd --distortions=false --variable_update=replicated --data_dir=/opt/imagenet/tf_train --data_name=imagenet --model=resnet50 --batch_size=32 --train_dir=/opt/data/resnet50_ckpt/cpu_v1/ --num_epochs=10--save_model_steps=20
Describe the expected behavior After training thousands batchs raise "tensorflow.python.framework.errors_impl.InvalidArgumentError: Retval[0] does not have value"
strace log about training: