Sachin Prasad

@danijar, Usually these versions are tested by Nvidia and the latest stable CUDA version against the specific Tensorflow version will be published.

Could you please provide some simple reproducible code and the error message which you are getting currently in order to expedite the troubleshooting process. Thanks!

Attaching the gist with reproducing error [here](https://colab.research.google.com/gist/sachinprasadhs/ae7f04984ab32dc6f7f0473f82491c32/44199.ipynb). The reported error can be avoided by registering the class with `@keras.utils.register_keras_serializable()` [here](https://colab.research.google.com/gist/sachinprasadhs/9471d05206aa1c206584c33a9d5d598f/44199.ipynb) is the working gist. However, this comment from the user...

You can use the solution mentioned [here](https://stackoverflow.com/questions/59864408/tensorflowyour-input-ran-out-of-data) to avoid warning and continue training.

@maifeeulasad , Could you please elaborate your requirement to understand the problem better. Thanks!

@ianstenbit , cc: @fchollet

Since the `max_total_size` could be any value of `int32`, it is not feasible to have a validation for that, considering the crash/OOM happens is based on the system workload. Details...

Hi, Could you please try with the nightly version, cublas issue seems to be fixed with the PR here https://github.com/tensorflow/tensorflow/pull/56691, but it is not available in 2.10.0.

You can try with the latest version, if it is working as intended, I would leave it as it is.

You can switch back to TensorRT 7.2.3.4 with tensorflow nightly version, until Tensorflow 2.11.0 is available, which would fix the dependency issues.