[Question] Running ScanNet in Docker on a MacBook with AMD chipset

agdiaz commented 8 months ago

Dear Jérôme, How are you? I hope you are doing well.

Abstract

I would like to try your software tool directly as a Docker container on my laptop that has an Apple Silico chipset (ARM architecture instead of x64). However, when I try to run it, I receive an error message:

Using TensorFlow backend.
Illegal instruction

Context information

My current environment is:

Laptop: Apple M2 Pro, OS version 14.2.1 (23C71).
Docker version: Docker version 24.0.7, build afdd53b
Test command: python predict_bindingsites.py 1brs_A --noMSA

Settings for Docker engine:

I have tested the image by doing:

$ DOCKER_DEFAULT_PLATFORM=linux/amd64 docker run --rm -it -v $PWD/data:/data jertubiana/scannet /bin/bash
root@2d0f05562a11:/ScanNet# python predict_bindingsites.py 1brs_A --noMSA
Using TensorFlow backend.
Illegal instruction

I have also cloned the repository and created another Dockerfile to install the dependencies (from the requirements.txt file) with the same results.

I have also found a comment on StackOverflow saying that downgrading the tensorflow version may fix it. However, it raised different errors due to invalid compatibility between the code and the libraries' versions.

Question(s)

Do you have a workaround to fix it or another docker image built for ARM architectures? Thank you so much in advance for your feedback and guidance.

Best regards from Brussels, Adrián

jertubiana commented 8 months ago

Hi Adrian, sorry for the delay, did you manage to fix the problem? One option could be to upgrade to Tensorflow2. Please email me at @.*** and I’ll send source code (still in dev.) Best, Jerome

On 22 Jan 2024, at 17:46, Adrián Diaz @.***> wrote:

Dear Jérôme, How are you? I hope you are doing well.

Abstract

I would like to try your software tool directly as a Docker container on my laptop that has an Apple Silico chipset (ARM architecture instead of x64). However, when I try to run it, I receive an error message:

Using TensorFlow backend. Illegal instruction Context information

My current environment is:

Laptop: Apple M2 Pro, OS version 14.2.1 (23C71). Docker version: Docker version 24.0.7, build afdd53b Test command: python predict_bindingsites.py 1brs_A --noMSA Settings for Docker engine: image.png (view on web) https://github.com/jertubiana/ScanNet/assets/1646576/cc9a3e99-688e-4f68-b5c5-60a9eb9128f9 I have tested the image by doing:

$ DOCKER_DEFAULT_PLATFORM=linux/amd64 docker run --rm -it -v $PWD/data:/data jertubiana/scannet /bin/bash @.***:/ScanNet# python predict_bindingsites.py 1brs_A --noMSA Using TensorFlow backend. Illegal instruction I have also cloned the repository and created another Dockerfile to install the dependencies (from the requirements.txt file) with the same results.

I have also found a comment on StackOverflow saying that downgrading the tensorflow version may fix it. However, it raised different errors due to invalid compatibility between the code and the libraries' versions.

Question(s)

Do you have a workaround to fix it or another docker image built for ARM architectures? Thank you so much in advance for your feedback and guidance.

Best regards from Brussels, Adrián

— Reply to this email directly, view it on GitHub https://github.com/jertubiana/ScanNet/issues/8, or unsubscribe https://github.com/notifications/unsubscribe-auth/ACHSPJQAALKDF2K2K53QOR3YP2CWJAVCNFSM6AAAAABCFN2RUSVHI2DSMVQWIX3LMV43ASLTON2WKOZSGA4TIMRQG44TMNQ. You are receiving this because you are subscribed to this thread.

agdiaz commented 8 months ago

Hello Jérôme,

Thank you for your answer. Indeed, the staff of the HPC at my university recommended to use TensorFlow2 in case I wanted to install it on their infrastructure, otherwise it can't be officially installed as a HPC module. I will contact you to see how I could help to test the dev version with TensorFlow2.

Kind regards, Adrián

jertubiana / ScanNet