Skip to content

feat(docker): provision ROUGE scorer deps and corpora in dev image #507

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
anandhu-eng wants to merge 1 commit into
base: arekay/push_image_workflow
Choose a base branch
from
Open

feat(docker): provision ROUGE scorer deps and corpora in dev image #507

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
30 changes: 28 additions & 2 deletions scripts/Dockerfile.dev
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -7,6 +7,10 @@
# evaluator, PROVISION_DSR1=1). It lives under examples/, not src/, so it is COPYed
# in explicitly. To skip it (leaner image, no wan22 accuracy), build with PROVISION_VBENCH=0:
# docker build -f scripts/Dockerfile.dev --build-arg PROVISION_VBENCH=0 --build-arg USER_ID=$(id -u) --build-arg GROUP_ID=$(id -g) -t inference-endpoint-dev .
#
# The ROUGE scorer's corpora + metric script are likewise prefetched by default
# (PROVISION_ROUGE=1) so `eval_method: "rouge"` needs no runtime network; build with
# PROVISION_ROUGE=0 to skip.

FROM python:3.12.11-slim

Expand Down Expand Up @@ -52,11 +56,11 @@ ENV PATH="/opt/venv/bin:/home/appuser/.local/bin:$PATH" \
VBENCH_CACHE_DIR=/opt/vbench_cache

# Install dependencies first (cached unless pyproject.toml/uv.lock change)
RUN uv sync --frozen --no-install-project --extra dev --extra test
RUN uv sync --frozen --no-install-project --extra dev --extra test --extra rouge

# Copy source and install project
COPY --chown=${USER_ID}:${GROUP_ID} src/ ./src/
RUN uv sync --frozen --extra dev --extra test
RUN uv sync --frozen --extra dev --extra test --extra rouge


# Provision the isolated DeepSeek-R1 accuracy evaluator. Gated by PROVISION_DSR1
Expand Down Expand Up @@ -103,6 +107,28 @@ RUN if [ "${PROVISION_VBENCH}" = "1" ]; then \
echo "PROVISION_VBENCH=${PROVISION_VBENCH}: skipping VBench provisioning" ; \
fi

# ROUGE scorer runtime data. The `rouge` extra above installs nltk/evaluate/rouge_score,
# but neither is self-contained: nltk.sent_tokenize needs the punkt/punkt_tab corpora and
# evaluate.load("rouge") fetches its metric script from the HF Hub. Without this block the
# image needs network at evaluation time for `eval_method: "rouge"` - the same gap
# prefetch_weights.py closes for VBench.
#
# Both land in their default per-user caches (~/nltk_data and ~/.cache/huggingface, i.e.
# appuser's home) and are baked into the image layer. Deliberately NOT redirected to /opt
# via NLTK_DATA/HF_HOME: HF_HOME is a global runtime env var, and pointing it elsewhere
# would orphan whatever the DSR1/VBench stages above already cached under the default path.
# The final python -c asserts the corpora actually resolve, so a silent download failure
# fails the build here rather than at evaluation time.
# PROVISION_ROUGE=0 skips this block (leaner image, no offline rouge accuracy).
ARG PROVISION_ROUGE=1
RUN if [ "${PROVISION_ROUGE}" = "1" ]; then \
python -c "import nltk; nltk.download('punkt'); nltk.download('punkt_tab')" && \
python -c "import evaluate; evaluate.load('rouge'); print('rouge metric cached')" && \
python -c "import nltk; assert nltk.sent_tokenize('One. Two.') == ['One.', 'Two.']; print('punkt corpora ok')" ; \
else \
echo "PROVISION_ROUGE=${PROVISION_ROUGE}: skipping ROUGE corpora/metric prefetch" ; \
fi

# OCI image metadata so `docker inspect` self-identifies this image (the endpoints benchmark
# client, distinct from the lcb-service evaluator image). Declared last so a wording change
# only rebuilds this tiny config layer, not the apt/uv/DSR1/VBench stages above.
Expand Down
Loading