How to find out keras model memory size?

konm123 · October 23, 2021, 11:35am

Hello!

I am doing a school work and I need to find out keras model memory size so I could compare different models. It is supposed to be composed of weights/parameters and model itself. It was given that model.summary() should contain all the information. From there I see that layer info and output shapes with the number of parameters.

I understand parameters as they are just a numbers and so, number of parameters * 4B most likely will give how much room parameters take. But I know that more information is required to know memory occupied by this model.

Can anyone guide me on how I can find out how many bytes of memory this model occupies?

Bhack · October 23, 2021, 12:37pm

You can retrieve memory info with flops from the profiler as in:

github.com/tensorflow/tensorflow

TF 2.0 Feature: Flops calculation

opened 09:54AM - 25 Sep 19 UTC

pzobel

stat:awaiting tensorflower type:feature comp:tfdbg

<em>Please make sure that this is a feature request. As per our [GitHub Policy](…https://github.com/tensorflow/tensorflow/blob/master/ISSUES.md), we only address code/doc bugs, performance issues, feature requests and build/installation issues on GitHub. tag:feature_template</em> **System information** - TensorFlow version (you are using): TF 2.0 RC2 - Are you willing to contribute it (Yes/No): **Describe the feature and the current behavior/state.** I am missing the opportunity to compute the number of floating point operations of a tf.keras Model in TF 2.0. In TF 1.x tf.profiler was available [see here](https://stackoverflow.com/questions/45085938) but I can find anything equivalent for TF 2.0 yet. **Will this change the current api? How?** **Who will benefit with this feature?** Everbody interested in the computational complexity of a TensorFlow model. **Any Other info.**

You can also try:

konm123 · October 23, 2021, 2:16pm

I am unable to figure out how the first link addresses my question. As for the second link, it tells me memory requirement, but not how much the model actually occupies memory. So I am still lost

Bhack · October 23, 2021, 2:27pm

The second link Is a good estimate of the on paper memory of your model.

The first one is the real memory in the runtime and e.g. you can retrieve It with ProfileOptionBuilder.time_and_memory() as tf.profiler option.

Other then the examples in the first link thread you can find more at:

github.com

tensorflow/tensorflow/blob/master/tensorflow/core/profiler/g3doc/python_api.md#time-and-memory

## Python API Tutorials

* [Parameters and Shapes](#parameters-and-shapes)
* [Float Operations](#float-operations)
* [Time and Memory](#time-and-memory)
* [Visualize](#visualize)
* [Multi-step Profiling](#multi-step-profiling)

```import tensorflow as tf```.

### Parameters and Shapes.
```python
# Print trainable variable parameter statistics to stdout.
ProfileOptionBuilder = tf.profiler.ProfileOptionBuilder

param_stats = tf.profiler.profile(
    tf.get_default_graph(),
    options=ProfileOptionBuilder.trainable_variables_parameter())

# Use code view to associate statistics with Python codes.

This file has been truncated. show original