char-fork/llama.cpp

git clone https://git.t4t.associates/char-fork/llama.cpp

Adrien Gallouëtdevops : add llama in all docker images (#25035)5d8ccdf9d

char/aliceai-t5-moe
brancheschar/aliceai-t5-moeall refs →
3.1 KiB124 linesraw
1ARG UBUNTU_VERSION=24.04
2ARG BUILD_DATE=N/A
3ARG APP_VERSION=N/A
4ARG APP_REVISION=N/A
5
6ARG NODE_VERSION=24
7
8FROM docker.io/node:$NODE_VERSION AS web
9
10ARG APP_VERSION
11
12WORKDIR /app/tools/ui
13
14COPY tools/ui/package.json tools/ui/package-lock.json ./
15RUN npm ci
16
17COPY tools/ui/ ./
18RUN LLAMA_BUILD_NUMBER="$APP_VERSION" npm run build
19
20FROM docker.io/ubuntu:$UBUNTU_VERSION AS build
21
22ARG TARGETARCH
23
24RUN apt-get update && \
25    apt-get install -y gcc-14 g++-14 build-essential git cmake libssl-dev
26
27ENV CC=gcc-14 CXX=g++-14
28
29WORKDIR /app
30
31COPY . .
32
33COPY --from=web /app/tools/ui/dist tools/ui/dist
34
35RUN if [ "$TARGETARCH" = "amd64" ] || [ "$TARGETARCH" = "arm64" ]; then \
36        cmake -S . -B build -DCMAKE_BUILD_TYPE=Release -DGGML_NATIVE=OFF -DLLAMA_BUILD_TESTS=OFF -DGGML_BACKEND_DL=ON -DGGML_CPU_ALL_VARIANTS=ON; \
37    else \
38        echo "Unsupported architecture"; \
39        exit 1; \
40    fi && \
41    cmake --build build -j $(nproc)
42
43RUN mkdir -p /app/lib && \
44    find build -name "*.so*" -exec cp -P {} /app/lib \;
45
46RUN mkdir -p /app/full \
47    && cp build/bin/* /app/full \
48    && cp *.py /app/full \
49    && cp -r conversion /app/full \
50    && cp -r gguf-py /app/full \
51    && cp -r requirements /app/full \
52    && cp requirements.txt /app/full \
53    && cp .devops/tools.sh /app/full/tools.sh
54
55## Base image
56FROM docker.io/ubuntu:$UBUNTU_VERSION AS base
57
58ARG BUILD_DATE=N/A
59ARG APP_VERSION=N/A
60ARG APP_REVISION=N/A
61ARG IMAGE_URL=https://github.com/ggml-org/llama.cpp
62ARG IMAGE_SOURCE=https://github.com/ggml-org/llama.cpp
63LABEL org.opencontainers.image.created=$BUILD_DATE \
64      org.opencontainers.image.version=$APP_VERSION \
65      org.opencontainers.image.revision=$APP_REVISION \
66      org.opencontainers.image.title="llama.cpp" \
67      org.opencontainers.image.description="LLM inference in C/C++" \
68      org.opencontainers.image.url=$IMAGE_URL \
69      org.opencontainers.image.source=$IMAGE_SOURCE
70
71RUN apt-get update \
72    && apt-get install -y libgomp1 curl ffmpeg \
73    && apt autoremove -y \
74    && apt clean -y \
75    && rm -rf /tmp/* /var/tmp/* \
76    && find /var/cache/apt/archives /var/lib/apt/lists -not -name lock -type f -delete \
77    && find /var/cache -type f -delete
78
79COPY --from=build /app/lib/ /app
80
81### Full
82FROM base AS full
83
84COPY --from=build /app/full /app
85
86WORKDIR /app
87
88RUN apt-get update \
89    && apt-get install -y \
90    git \
91    python3 \
92    python3-pip \
93    python3-wheel \
94    && pip install --break-system-packages --upgrade setuptools \
95    && pip install --break-system-packages -r requirements.txt \
96    && apt autoremove -y \
97    && apt clean -y \
98    && rm -rf /tmp/* /var/tmp/* \
99    && find /var/cache/apt/archives /var/lib/apt/lists -not -name lock -type f -delete \
100    && find /var/cache -type f -delete
101
102ENTRYPOINT ["/app/tools.sh"]
103
104### Light, CLI only
105FROM base AS light
106
107COPY --from=build /app/full/llama /app/full/llama-cli /app/full/llama-completion /app
108
109WORKDIR /app
110
111ENTRYPOINT [ "/app/llama-cli" ]
112
113### Server, Server only
114FROM base AS server
115
116ENV LLAMA_ARG_HOST=0.0.0.0
117
118COPY --from=build /app/full/llama /app/full/llama-server /app
119
120WORKDIR /app
121
122HEALTHCHECK CMD [ "curl", "-f", "http://localhost:8080/health" ]
123
124ENTRYPOINT [ "/app/llama-server" ]