Dockerfile

# Layer 1: Build the virtual environment
FROM ghcr.io/lambda-feedback/baseevalutionfunctionlayer:3.11 AS builder

RUN pip install poetry==1.8.3

ENV POETRY_NO_INTERACTION=1 \
    POETRY_VIRTUALENVS_IN_PROJECT=1 \
    POETRY_VIRTUALENVS_CREATE=1 \
    POETRY_CACHE_DIR=/tmp/poetry_cache

COPY pyproject.toml ./
RUN poetry lock

RUN --mount=type=cache,target=$POETRY_CACHE_DIR \
    poetry install --without dev --no-root

# Layer 2: Download NLTK models
FROM ghcr.io/lambda-feedback/baseevalutionfunctionlayer:3.11 AS models

ENV NLTK_DATA=/usr/share/nltk_data

ENV MODEL_PATH=/app/app/models

RUN yum install -y \
    wget \
    unzip

COPY ./scripts ./scripts

RUN ./scripts/download_models.sh
RUN ./scripts/download_nltk.sh

# Layer 3: Final image
FROM ghcr.io/lambda-feedback/baseevalutionfunctionlayer:3.11

ENV VIRTUAL_ENV=/app/.venv \
    PATH="/app/.venv/bin:$PATH"

ENV NLTK_DATA=/usr/share/nltk_data

ENV MODEL_PATH=/app/app/models

# Copy the evaluation function to the app directory
COPY ./evaluation_function ./app

# Precompile python files for faster startup
RUN python -m compileall -q .

COPY --from=builder ${VIRTUAL_ENV} ${VIRTUAL_ENV}
COPY --from=models ${NLTK_DATA} ${NLTK_DATA}
COPY --from=models ${MODEL_PATH} ${MODEL_PATH}

RUN export PYTHONPATH=$PYTHONPATH:/app/app

# Set permissions so files and directories can be accessed on AWS
RUN chmod 644 $(find ./app/ -type f)
RUN chmod 755 $(find ./app/ -type d)

# The entrypoint for AWS is to invoke the handler function within the app package
CMD [ "/app/app.handler" ]