Session crashed after trying to decompress the tfci image in Colab

[Su Zi Wei]

Hi,

I have been trying out the Colab code: code https://colab.research.google.com/github/tensorflow/compression/blob/master/models/hific/colab.ipynb

And if I run everything with default setting (hific-lo, default image etc.), the Colab will crash at the tfci.decompress step in the 'Compress images' section, saying that RAM exceeded.

However, this did not happen before when I ran the same code on Colab and only appear to happen recently. I am wondering if this is an issue caused by Colab or the code itself? I have also tried to run tfci locally on my laptop and the same issue occurs. Any help is appreciated.

`Regards,

Ziwei

[Fabian Mentzer]

Hi Ziwei

Thanks for reaching out. I can reproduce your error when running the colab. I am not sure what the exact error is, all I see is the following lines in the colab log

> 2022-11-14 07:08:55.959279: W tensorflow/core/common_runtime/gpu/gpu_bfc_allocator.cc:42] Overriding orig_value setting because the TF_FORCE_GPU_ALLOW_GROWTH environment variable is set. Original config value was 0.

> KernelRestarter: restarting kernel (1/5), keep random ports

> WARNING:root:kernel 3f50bf0b-e4e7-467b-bdec-582922787be0 restarted

According to what the cell prints, it happens while decompressing

+Johannes Ballé Are you aware of any recent RAM issues with TFCI?

@Ziwei what do you mean wiht  " have also tried to run tfci locally on my laptop and the same issue occurs."? Does the script simply crash without error also locally?

--
You received this message because you are subscribed to the Google Groups "tensorflow-compression" group.
To unsubscribe from this group and stop receiving emails from it, send an email to tensorflow-compre...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/tensorflow-compression/7c8e7df7-f95b-4666-ab5c-f5806ed0edf8n%40googlegroups.com.

[Johannes Ballé]

Hi,

I'm not aware of any recent RAM issues. Have you tried going back to the previous TF/TFC version (2.9.x), to see if the problem happens there as well?

Johannes