Kubeflow fastai library Setup

zlapp · March 31, 2019, 11:57am

Hi,

Kubeflow seems like a great way to develop, serve, and maintain ml models.

I was wondering if anyone could share their experience/best practices for setting Kubeflow with the fastai library. Is there a docker image available? Is it simple to use?

ptrampert · April 1, 2019, 2:39pm

I use fastai with Kubeflow and Docker. You have to build your own container. Works like charm.

zlapp · April 1, 2019, 4:25pm

Thanks @ptrampert could you share a dockerfile as a reference? I tried using the official kaggle dockerfile which includes many libraries including fastai but it wouldn’t load correctly.

sariabod · April 1, 2019, 10:10pm

hello @zlapp, here is my Docker file I use to deploy FastAI apps. Maybe you can use it as a starting point.

github.com

sariabod/fastai_docker/blob/master/app/Dockerfile

FROM nvidia/cuda:10.0-base-ubuntu18.04

# Install some basic utilities
RUN apt-get update && apt-get install -y \
    curl \
    ca-certificates \
    sudo \
    git \
    bzip2 \
    libx11-6 \
 && rm -rf /var/lib/apt/lists/*

# Create a working directory
RUN mkdir /app
ADD . /app
WORKDIR /app

# Install Miniconda
RUN curl -so ~/miniconda.sh https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh && chmod +x ~/miniconda.sh && ~/miniconda.sh -b -p ~/miniconda && rm ~/miniconda.sh
ENV PATH=/root/miniconda/bin:$PATH

This file has been truncated. show original

To use GPUs you will need nvidia-docker2 otherwise you can just launch without the Nvidia runtime to use the CPU for inference.

ptrampert · April 2, 2019, 4:17am

Try here: