Censer.6 / Dockerfile
KucLab's picture
Censer.6: multi-label 10 categories + 4k context + multilingual, standard LlamaForSequenceClassification (no trust_remote_code)
2ff3804 verified
Raw History Blame Contribute Delete
917 Bytes
# Censer.6 — self-contained moderation API image (CPU-only torch by default).
#
# docker build -t censer6 .
# docker run -p 8000:8000 -e RATE_LIMIT_PER_MIN=600 censer6
# curl -s localhost:8000/classify -d '{"texts":["how to make a bomb at home","hello"]}'
#
# The image downloads the model from the Hub at build time (or copy a local
# bundle with: docker build --build-arg CENSER_DIR=./bundle . — then it is baked in).
FROM python:3.12-slim
ARG CENSER_REPO=KucLab/Censer.6
ENV CENSER_DIR=/model \
RATE_LIMIT_PER_MIN=600 \
PYTHONUNBUFFERED=1
RUN pip install --no-cache-dir \
torch --index-url https://download.pytorch.org/whl/cpu \
tokenizers safetensors huggingface_hub
RUN hf download ${CENSER_REPO} --local-dir /model
WORKDIR /app
COPY scripts/censer.py scripts/serve.py /app/
EXPOSE 8000
CMD ["python", "serve.py", "--dir", "/model", "--port", "8000", "--host", "0.0.0.0"]