2021-01-10 02:43

UserWarning: You are not using the TensorFlow backend

I'm filing an issue on behalf of Wolfgang Kerzendorf - the code is his, not mine. I've begun to debug this issue and here's what I've found so far.


When the code is run, there is a UserWarning that pops up saying "You are not using the TensorFlow Backend."

INFO:plaidml:Opening device "opencl_amd_radeon_pro_560x_compute_engine.0"
UserWarning: You are not using the TensorFlow backend. write_graph was set to False
  warnings.warn('You are not using the TensorFlow backend. '

The program runs through model.compile before the warning appears.

Replication of the error




  • 点赞
  • 写回答
  • 关注问题
  • 收藏
  • 复制链接分享
  • 邀请回答


  • weixin_39949506 weixin_39949506 4月前


    点赞 评论 复制链接分享
  • weixin_39837867 weixin_39837867 4月前

    Hi ,

    Looking further into this issue, I've traced the warning back to the usage of TensorBoard.

    Currently, the code implementing TensorBoard on the Keras side checks explicitly for a TensorFlow backend. While this could be worked around if it was the only issue, I also noticed that you had tried to use the write_graph parameter, indicating that you want to use graph visualizations within TensorBoard. Currently, graph visualization features are not available in the PlaidML backend.

    As for your broader question:

    It seems that [PlaidML] doesn't work with the new version of Keras that is bundled with Tensorflow and I was wondering if you guys are supporting this in the near future?

    The version of Keras that's bundled with TensorFlow would need to use PlaidML bundled with nGraph to utilize nGraph's TensorFlow support. You can use the ngraph-tf bridge to connect nGraph with PlaidML. Please let us know if/when you try this and what your results are.

    Thanks for using PlaidML!

    点赞 评论 复制链接分享
  • weixin_39541044 weixin_39541044 4月前

    I am deploying a Tensorflow/Keras application. I would like to speed up this application and not switch to standard Keras. Is there a way to install PlaidML so that it ties into the Tensorflow Keras source instead?

    点赞 评论 复制链接分享
  • weixin_39818521 weixin_39818521 4月前

    any plans tensorboard graph will work with plaidml?

    点赞 评论 复制链接分享