Question posted in
The official documentation can be found here.

Unable to run tensorflow in Docker

I have created a Dockerfile as shown below:

FROM tensorflow/tensorflow:latest

WORKDIR /code

COPY ./requirements.txt /code/requirements.txt

RUN pip install --no-cache-dir --upgrade -r /code/requirements.txt

COPY ./app /code/app

CMD ["fastapi", "run", "app/main.py", "--port", "80"]

and requirements.txt file with following values:

fastapi>=0.110.0,<0.113.0
pydantic>=2.7.0,<3.0.0
tensorflow

I am using the following command to build the docker and run it then

docker build --platform=linux/arm64/v8 -t myimage .
docker run -d --name mycontainer -p 80:80 myimage

and main function has following code:

from typing import Union

from fastapi import FastAPI, Request

# from tensorflow import keras
# import tensorflow
# model = keras.models.load_model("app/my_model.keras")

app = FastAPI()


@app.get("/")
def read_root():
    return {"Hello": "World"}


@app.get("/items/{item_id}")
def read_item(item_id: int, q: Union[str, None] = None):
    return {"item_id": item_id, "q": q}

Whenever i try to import tensorflow in above main file there is an error coming in Docker run. Any resolution for the same

I tried changing the From path to python 3.9 but it was causing compilation error while installing tensorflow.

Tags:

2

Answers


  1. 0

    If you may, try with an image that’s a better match.

    FROM tensorflow/tensorflow:2.9.0

    Check for the best match for ARM and the dependencies.

    Login or Signup to reply.
  2. 0

    if you are running docker on Mac M1 or later versions you won’t be able to use prebuild docker images. At least as of Today.

    I have Mac Por M1. I have checked several versions of prebuilt docker images:

    Tensorflow 2.9 version:

    docker run -it --rm tensorflow/tensorflow:2.9.0 
# next line inside the container
python -c "import tensorflow"

    This will give an error:

    The TensorFlow library was compiled to use AVX instructions, but these aren't available on your machine.
Aborted

    Tensorflow 2.16 (latest as of Today) version.

    # v2.16
docker run -it --rm tensorflow/tensorflow:latest bash  
# next line inside the container
python -c "import tensorflow"

    This simply outputs

    Illegal instruction

    I have also testes prebuilt images with versions between 2.9 and 2.16 and neither of them works on the Mac. However this was expected, I guess…

    Thus I would conclude that the only option is to compile custom docker tensorflow image that supports mac with ARM64.

    Login or Signup to reply.
Please signup or login to give your own answer.

Back To Top
Search