search

IBM / tensorflow-large-model-support

Large Model Support in Tensorflow
Apache License 2.0
202 stars 38 forks source link

ImportError: cannot import name 'LMSKerasCallback' #16

Closed ticlazau closed 5 years ago

ticlazau commented 5 years ago

Hi Sam,

I am using WML-CE 1.6.1 (TF 1.14) with a provided synthetic example: Keras_ResNet50.py

$python Keras_ResNet50.py --image_size 2300 2019-06-24 04:10:47.551637: I tensorflow/stream_executor/platform/default/dso_loader.cc:42] Successfully opened dynamic library libcudart.so.10.1 WARNING: Logging before flag parsing goes to stderr. W0624 04:10:49.657887 140735954111392 deprecation_wrapper.py:119] From /home/coredump/anaconda3/envs/powerai-1.6.1/lib/python3.6/site-packages/tflms-2.0.1-py3.6.egg/tensorflow_large_model_support/lms.py:25: The name tf.train.SessionRunHook is deprecated. Please use tf.estimator.SessionRunHook instead.

Traceback (most recent call last): File "Keras_ResNet50.py", line 60, in from tensorflow_large_model_support import LMSKerasCallback ImportError: cannot import name 'LMSKerasCallback'

Any advice?

smatzek commented 5 years ago

It appears you are attempting to use the Keras_ResNet50.py example from this github repository with WML CE 1.6.1. As noted in this repository's readme, the Large Model Support in PowerAI (and WML CE) is newer implementation. WML CE's new implementation is not API compatible with the version in this repository.

The PowerAI / WML CE documentation points at the https://github.com/IBM/powerai/ repository for its examples. The repository has branches for the different PowerAI/WML CE releases. The TFLMSv2 (new implementation examples) for WML CE 1.6.1 are here: https://github.com/IBM/powerai/tree/wmlce-1.6.1/examples/tensorflow_large_model_support/v2