Dockerfile

[vishnubob]

Thanks for publishing this work. Here is a Dockerfile that builds a compatible runtime environment for executing this model.

FROM nvidia/cuda:10.0-cudnn7-runtime-ubuntu18.04
RUN apt-get update && \
    apt-get install -y python3-pip && \
    pip3 install \
    numpy==1.16.3 \
    tensorflow-gpu==1.13.1 \
    imageio==2.8.0 \
    fire>=0.1.3 \
    regex==2017.4.5 \
    requests==2.21.0 \
    tqdm==4.31.1 \
    scipy==1.4.1

Clone this repo, and drop this Dockerfile in the root of the repo. Build the container:

docker build -t image-gpt .

With the root of repo as your current directory, run the container:

docker run -it -v $(pwd):/app -v $HOME/image-gpt-data:/root image-gpt bash

[asisakov]

Perhaps it will be better to use a lower amount of memory and initially set:
FROM python:3.6-stretch
Not Ubuntu.

Also, we can use:

COPY requirements.txt .
RUN pip install --no-cache-dir -r requirements.txt

Instead of RUN ....

[vishnubob]

Perhaps it will be better to use a lower amount of memory and initially set:
...
Also, we can use:
COPY requirements.txt .
RUN pip install --no-cache-dir -r requirements.txt

Thanks for your suggestions. Ubuntu or Debian aside, the image I picked includes the right CUDA and CUDNN versions necessary to run this work on a GPU. At the time I wrote and posted this Dockerfile, there were missing dependencies in the requirements.txt, so that's why they are hard coded. The intent was not to provide something canonical or ideal, it was to help people get started during this early phase of publishing. So, after testing this Dockerfile to make sure it works, here it is.

Can I ask, did you test your suggestions?

[asisakov]

includes the right CUDA and CUDNN versions necessary to run this work on a GPU

there were missing dependencies

Thank you for sharing. I didn't notice it at first.

I did test my code, but not the whole. When data started to download, in my thoughts, it worked well. But my mistake is that I didn't try to run the model. Maybe the process would show me the incomplete (in your words) requirements.
If you are interested in my code, you can check it in pull-requests.

[asisakov]

Can I add your part with Ubuntu to Dockerfile?

[vishnubob]

Please, add away!

[yugo-harago]

I'm having this error:
"ImportError: libcuda.so.1: cannot open shared object file: No such file or directory"
When I run "python src/run.py --sample --n_embd 512 --n_head 8 --n_layer 24"