How to add report_tensor_allocations_upon_oom to RunOptions in Keras

Question

I'm trying to train a neural net on a GPU using Keras and am getting a "Resource exhausted: OOM when allocating tensor" error. The specific tensor it's trying to allocate isn't very big, so I assume some previous tensor consumed almost all the VRAM. The error message comes with a hint that suggests this:

Hint: If you want to see a list of allocated tensors when OOM happens, add report_tensor_allocations_upon_oom to RunOptions for current allocation info.

That sounds good, but how do I do it? RunOptions appears to be a Tensorflow thing, and what little documentation I can find for it associates it with a "session". I'm using Keras, so Tensorflow is hidden under a layer of abstraction and its sessions under another layer below that.

How do I dig underneath everything to set this option in such a way that it will take effect?

Solution 1

TF1 solution:

Its not as hard as it seems, what you need to know is that according to the documentation, the **kwargs parameter passed to model.compile will be passed to session.run

So you can do something like:

import tensorflow as tf
run_opts = tf.RunOptions(report_tensor_allocations_upon_oom = True)

model.compile(loss = "...", optimizer = "...", metrics = "..", options = run_opts)

And it should be passed directly each time session.run is called.

TF2:

The solution above works only for tf1. For tf2, unfortunately, it appears there is no easy solution yet.

Solution 2

Currently, it is not possible to add the options to model.compile. See: https://github.com/tensorflow/tensorflow/issues/19911

Solution 3

OOM means out of memory. May be it is using more memory at that time. Decrease batch_size significantly. I set to 16, then it worked fine

Comments

[dspeyer]

I used options=run_opts, since it's a kwargs thing, and that worked

[Amila]

@Matias Valdenegro I get ValueError: ('Some keys in session_kwargs are not supported at this time: %s', dict_keys(['options'])). Any idea what I'm doing wrong?

[Enea Dume]

While this link may answer the question, it is better to include the essential parts of the answer here and provide the link for reference. Link-only answers can become invalid if the linked page changes.

[zwep]

I had this exact same issue. Using keras version 2.2.4... is there any solution?

[zwep]

Also, I received this error 'Protocol message RunOptions has no "report_tensor_allocations_upon_oom" field.'

[Dan Grahn]

@zwep this was resolved in 2.2.4. Are you sure you've updated?

[Zaccharie Ramzi]

This caused me a segmentation fault for some reason: [1] 3957 segmentation fault python oom_net.py

[Zaccharie Ramzi]

Apparently I was not the only one with a segfault: github.com/keras-team/keras/issues/11322

[Adam Azarchs]

Whether that will work, and what batch size is appropriate, will depend entirely on the model in question, as well as the dataset. If one is attempting to debug a memory issue that doesn't depend on batch size, this doesn't help at all.