How do I install libtensorflow for the Recognize app to run natively?

[JL102]

Hi,

The Recognize app expects libtensorflow to be installed in order to run natively, but it seems that that would require modifying the nextcloud-aio-nextcloud container. I got instructions from ChatGPT to clone the nextcloud AIO image and install libtensorflow in it, but I don't think it'll work because libtensorflow will be added to the master container, not the nextcloud container - Right?

FROM nextcloud/all-in-one:latest

# Install dependencies for libtensorflow
RUN apt-get update \
    && apt-get install -y libtensorflow1

I was instructed to then run docker build -t nextcloud-with-tensorflow . and then replace nextcloud/all-in-one:latest in my docker run command with nextcloud-with-tensorflow`.

I'm hesitant to test it because I'm really not familiar with Docker and I'm afraid of messing something up. I'd rather ask for assistance. Would this technique work, or would libtensorflow only be installed in the mastercontainer, leading to Recognize not having access to it? If it's the latter, then how do I do it properly?

[szaimen]

Hi, libtensorflow does unfortunately not work with native speed inside an Alpine based container.

However I would gladly add an additional container for this after nextcloud/recognize#73 is implemented.

[Zoey2936]

just for me, why does it not run with full speed?

[szaimen]

Because it needs glibc for that.

[Zoey2936]

can you maybe give me links where this is written, please?

[JL102]

@szaimen Correct me if I'm wrong but I think it's simply because the AIO docker image doesn't contain the necessary libraries. If someone really wanted to, they could fork the image and make their own version with the necessary dependencies, right?

[szaimen]

tensorflow is simply incompatible with alpine linux. See #1569 (comment) and tensorflow/tfjs#1425. Because of that only WASM mode is poasible