Can't build Docker image for Ubuntu 20.04

[danielvachalek]

Tried following these instructions: https://lambdalabs.com/blog/set-up-a-tensorflow-gpu-docker-container-using-lambda-stack-dockerfile/

When I run this step

Build a docker image named lambda-stack:20.04

sudo docker build -t lambda-stack:20.04 -f Dockerfile.focal git://github.com/lambdal/lambda-stack-dockerfiles.git

I get an error message Errors were encountered while processing: /tmp/apt-dpkg-install-vKTQpk/001-libcudnn7_7.6.5-0lambda1_amd64.deb /tmp/apt-dpkg-install-vKTQpk/032-libcudnn-dev_7.6.5-0lambda1_amd64.deb

E: Sub-process /usr/bin/dpkg returned an error code (1) The command '/bin/sh -c apt-get update && apt-get install --yes gnupg && apt-key add lambda.gpg && rm lambda.gpg && echo "deb http://archive.lambdalabs.com/ubuntu focal main" > /etc/apt/sources.list.d/lambda.list && echo "Package: " > /etc/apt/preferences.d/lambda && echo "Pin: origin archive.lambdalabs.com" >> /etc/apt/preferences.d/lambda && echo "Pin-Priority: 1001" >> /etc/apt/preferences.d/lambda && echo "cudnn cudnn/license_preseed select ACCEPT" | debconf-set-selections && apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install --yes --no-install-recommends --option "Acquire::http::No-Cache=true" --option "Acquire::http::Pipeline-Depth=0" lambda-stack-cuda lambda-server && rm -rf /var/lib/apt/lists/' returned a non-zero code: 100

[picekl]

Hi Daniel,

I have the same or similar issue. From my experience, there is no working dockerfile offered here. Or a solution to make it build.

Try to use this image. -- > https://ngc.nvidia.com/catalog/containers/nvidia:tensorflow/tags

Works for me but the performance on the 3080 is quite strange.

Best, Lukas

[danielvachalek]

Thanks, Lukas! I'll give those images a try.

[simplehx]

I have the same problem.

[picekl]

Thanks, Lukas! I'll give those images a try.

Let me know if it worked for you. Here are some results from my side (will update it later with more experiments).

2080S vs 3080:

• MobileNet V2 - 10% speed increase.

[picekl]

I have the same problem.

Did you try the official nvidia image?

[danielvachalek]

@picekl Thanks again, I was able to get the Nvidia image to run. Now working through linking this image with the fast.ai to make a ready-made fast.ai container. Edit: clarity

[picekl]

@danielvachalek Are you talking about the Lambda or Nvidia image?

[sclarkson]

Could you try the most recent version? 30bcae0

[picekl]

Could you try the most recent version? 30bcae0

@sclarkson Thanks. This helped. FYI the speed up is the same as for the Nvidia Image. I will try another architecture soon.

[danielvachalek]

I was able to build the docker image and follow the tutorial, thanks all.