automated-codejson-generator/Dockerfile.base at dev · DSACMS/automated-codejson-generator · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
# base image that holds everything heavy and slow-changing (node-llama-cpp, the model, scc)
# pushed to GHCR by hand, the thin Dockerfile builds the action code on top of it
# see "Rebuilding the base image" in CONTRIBUTING.md for when and how

ARG MODEL_REPO=unsloth/gemma-4-E2B-it-GGUF
ARG MODEL_FILE=gemma-4-E2B-it-Q4_K_M.gguf
ARG MODEL_REVISION=0314792d7f1f7e229411f620751375812bb9faf2
ARG MODEL_SHA256=740185b21d22ceb83a11c3aa62ad5842ef32c70f6096d756bbee85a1e4ec34b8

# must match the version package-lock.json resolves, the thin Dockerfile checks this
ARG NODE_LLAMA_CPP_VERSION=3.21.1

# stage 1: model
# download the model GGUF and cache
FROM debian:bookworm-slim AS model
ARG MODEL_REPO
ARG MODEL_FILE
ARG MODEL_REVISION
ARG MODEL_SHA256

RUN apt-get update && apt-get install -y --no-install-recommends \
    ca-certificates \
    curl \
    && rm -rf /var/lib/apt/lists/*

RUN mkdir -p /opt/models \
    && curl -fSL --retry 3 \
        "https://huggingface.co/${MODEL_REPO}/resolve/${MODEL_REVISION}/${MODEL_FILE}?download=true" \
        -o /opt/models/model.gguf \
    && echo "${MODEL_SHA256}  /opt/models/model.gguf" | sha256sum -c -

# stage 2: base
FROM node:20-bookworm-slim AS base
ARG MODEL_FILE
ARG MODEL_REVISION
ARG NODE_LLAMA_CPP_VERSION

LABEL org.opencontainers.image.source=https://github.com/DSACMS/automated-codejson-generator \
      org.opencontainers.image.description="Base image for automated-codejson-generator: node-llama-cpp, the model and scc" \
      gov.cms.acg.node-llama-cpp.version="${NODE_LLAMA_CPP_VERSION}" \
      gov.cms.acg.model.file="${MODEL_FILE}" \
      gov.cms.acg.model.revision="${MODEL_REVISION}"

RUN apt-get update && apt-get install -y --no-install-recommends \
        git \
        ca-certificates \
        curl \
    && rm -rf /var/lib/apt/lists/*

# scc used for the laborHours estimate
RUN curl -fSL --retry 3 \
        "https://github.com/boyter/scc/releases/download/v3.4.0/scc_Linux_x86_64.tar.gz" \
        -o /tmp/scc.tar.gz \
    && echo "8f812d25c237112ea74a4285f0699b7a2c448d70c5e63e370d2d199d38b13eda  /tmp/scc.tar.gz" | sha256sum -c - \
    && tar -xzf /tmp/scc.tar.gz -C /usr/local/bin scc \
    && rm /tmp/scc.tar.gz \
    && chmod +x /usr/local/bin/scc \
    && scc --version

WORKDIR /app

# node-llama-cpp is the only runtime dependency rollup leaves out of dist/index.js
# the prebuilt linux-x64 binaries come in as optional deps, so skip the source download
# the thin image copies its own package.json over this placeholder one
ENV NODE_LLAMA_CPP_SKIP_DOWNLOAD=true
RUN echo '{"private": true}' > package.json \
    && npm install --omit=dev --no-audit --no-fund --no-package-lock \
        "node-llama-cpp@${NODE_LLAMA_CPP_VERSION}" \
    && npm cache clean --force \
    && test "$(node -p "require('/app/node_modules/node-llama-cpp/package.json').version")" = "${NODE_LLAMA_CPP_VERSION}"

COPY --from=model /opt/models/model.gguf /opt/models/model.gguf

ENV ACG_MODEL_PATH=/opt/models/model.gguf