docker.gpt-2-simple

FROM tensorflow/tensorflow:1.13.0rc0-gpu-py3

RUN apt-get update

RUN apt-get install \
	vim \
	less \
	curl \
	wget \
	pigz -y

RUN mkdir -p /opt

COPY ./gpt-2-simple /opt/gpt-2-simple


WORKDIR /opt/gpt-2-simple

# install custom gpt-2-simple package

RUN pip install -r requirements.txt
RUN pip install -e .


# install unpacking script
# call it using "# extract file.extension"
COPY ./extract.sh /opt/extract.sh


ENV WORKSPACE /workspace
RUN mkdir -p $WORKSPACE
WORKDIR $WORKSPACE

# training wrapper
COPY ./train_language.sh /opt/train_language.sh

# INRIA OSCAR Dataset https://traces1.inria.fr/oscar/
ENV DATASET_URL https://traces1.inria.fr/oscar/files/Compressed/

# set default language learning
ENV LANGUAGE en

# DATASET first $DATASET_SIZE MB for the desired language
# this is to avoid a too big dataset
ENV DATASET_SIZE 10M

#PARAMS for training
ENV MODEL_NAME 124M
ENV CHECKPOINT_DIR $WORKSPACE/checkpoint/$RUN_NAME
ENV MODEL_DIR $WORKSPACE/models
ENV DATASET $WORKSPACE/dataset
ENV STEPS 5000
ENV RESTORE_FROM latest
ENV SAMPLE_EVERY 1000
ENV SAVE_EVERY 1000
ENV PRINT_EVERY 1000
ENV OPTIMIZER adam
ENV OVERWRITE false
ENV MULTI_GPU true
ENV REPORTING_CONFIG $WORKSPACE/config.yml

ENTRYPOINT ["/bin/bash"]

CMD ["/opt/train_language.sh"]