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 →
5.7 KiB163 linesraw
1# ==============================================================================
2# ARGUMENTS
3# ==============================================================================
4
5# Define the CANN base image for easier version updates later
6ARG CHIP_TYPE=910b
7ARG CANN_BASE_IMAGE=quay.io/ascend/cann:8.5.0-${CHIP_TYPE}-openeuler24.03-py3.11
8ARG BUILD_DATE=N/A
9ARG APP_VERSION=N/A
10ARG APP_REVISION=N/A
11
12# ==============================================================================
13# BUILD STAGE
14# Compile all binary files and libraries
15# ==============================================================================
16ARG NODE_VERSION=24
17
18FROM docker.io/node:$NODE_VERSION AS web
19
20ARG APP_VERSION
21
22WORKDIR /app/tools/ui
23
24COPY tools/ui/package.json tools/ui/package-lock.json ./
25RUN npm ci
26
27COPY tools/ui/ ./
28RUN LLAMA_BUILD_NUMBER="$APP_VERSION" npm run build
29
30FROM ${CANN_BASE_IMAGE} AS build
31
32# -- Install build dependencies --
33RUN yum install -y gcc g++ cmake make git openssl-devel python3 python3-pip && \
34    yum clean all && \
35    rm -rf /var/cache/yum
36
37# -- Set the working directory --
38WORKDIR /app
39
40# -- Copy project files --
41COPY . .
42
43COPY --from=web /app/tools/ui/dist tools/ui/dist
44
45# -- Set CANN environment variables (required for compilation) --
46# Using ENV instead of `source` allows environment variables to persist across the entire image layer
47ENV ASCEND_TOOLKIT_HOME=/usr/local/Ascend/ascend-toolkit/latest
48ENV LD_LIBRARY_PATH=${ASCEND_TOOLKIT_HOME}/lib64:${LD_LIBRARY_PATH}
49ENV PATH=${ASCEND_TOOLKIT_HOME}/bin:${PATH}
50ENV ASCEND_OPP_PATH=${ASCEND_TOOLKIT_HOME}/opp
51ENV LD_LIBRARY_PATH=${ASCEND_TOOLKIT_HOME}/runtime/lib64/stub:$LD_LIBRARY_PATH
52# ... You can add other environment variables from the original file as needed ...
53# For brevity, only core variables are listed here. You can paste the original ENV list here.
54
55# -- Build llama.cpp --
56# Use the passed CHIP_TYPE argument and add general build options
57ARG CHIP_TYPE
58RUN source /usr/local/Ascend/ascend-toolkit/set_env.sh --force \
59    && \
60    cmake -B build \
61        -DGGML_CANN=ON \
62        -DCMAKE_BUILD_TYPE=Release \
63        -DSOC_TYPE=ascend${CHIP_TYPE} \
64        -DUSE_ACL_GRAPH=ON \
65        . && \
66    cmake --build build --config Release -j$(nproc)
67
68# -- Organize build artifacts for copying in later stages --
69# Create a lib directory to store all .so files
70RUN mkdir -p /app/lib && \
71    find build -name "*.so*" -exec cp -P {} /app/lib \;
72
73# Create a full directory to store all executables and Python scripts
74RUN mkdir -p /app/full && \
75    cp build/bin/* /app/full/ && \
76    cp *.py /app/full/ && \
77    cp -r conversion /app/full/ && \
78    cp -r gguf-py /app/full/ && \
79    cp -r requirements /app/full/ && \
80    cp requirements.txt /app/full/
81    # If you have a tools.sh script, make sure it is copied here
82    # cp .devops/tools.sh /app/full/tools.sh
83
84# ==============================================================================
85# BASE STAGE
86# Create a minimal base image with CANN runtime and common libraries
87# ==============================================================================
88FROM ${CANN_BASE_IMAGE} AS base
89
90ARG BUILD_DATE=N/A
91ARG APP_VERSION=N/A
92ARG APP_REVISION=N/A
93ARG IMAGE_URL=https://github.com/ggml-org/llama.cpp
94ARG IMAGE_SOURCE=https://github.com/ggml-org/llama.cpp
95LABEL org.opencontainers.image.created=$BUILD_DATE \
96      org.opencontainers.image.version=$APP_VERSION \
97      org.opencontainers.image.revision=$APP_REVISION \
98      org.opencontainers.image.title="llama.cpp" \
99      org.opencontainers.image.description="LLM inference in C/C++" \
100      org.opencontainers.image.url=$IMAGE_URL \
101      org.opencontainers.image.source=$IMAGE_SOURCE
102
103# -- Install runtime dependencies --
104RUN yum install -y libgomp curl && \
105    yum clean all && \
106    rm -rf /var/cache/yum
107
108# -- Set CANN environment variables (required for runtime) --
109ENV ASCEND_TOOLKIT_HOME=/usr/local/Ascend/ascend-toolkit/latest
110ENV LD_LIBRARY_PATH=/app:${ASCEND_TOOLKIT_HOME}/lib64:${LD_LIBRARY_PATH}
111ENV PATH=${ASCEND_TOOLKIT_HOME}/bin:${PATH}
112ENV ASCEND_OPP_PATH=${ASCEND_TOOLKIT_HOME}/opp
113# ... You can add other environment variables from the original file as needed ...
114
115WORKDIR /app
116
117# Copy compiled .so files from the build stage
118COPY --from=build /app/lib/ /app
119
120# ==============================================================================
121# FINAL STAGES (TARGETS)
122# ==============================================================================
123
124### Target: full
125# Complete image with all tools, Python bindings, and dependencies
126# ==============================================================================
127FROM base AS full
128
129COPY --from=build /app/full /app
130
131# Install Python dependencies
132RUN yum install -y git python3 python3-pip && \
133    pip3 install --no-cache-dir --upgrade pip setuptools wheel && \
134    pip3 install --no-cache-dir -r requirements.txt && \
135    yum clean all && \
136    rm -rf /var/cache/yum
137
138# You need to provide a tools.sh script as the entrypoint
139ENTRYPOINT ["/app/tools.sh"]
140# If there is no tools.sh, you can set the default to start the server
141# ENTRYPOINT ["/app/llama-server"]
142
143### Target: light
144# Lightweight image containing only llama-cli and llama-completion
145# ==============================================================================
146FROM base AS light
147
148COPY --from=build /app/full/llama /app/full/llama-cli /app/full/llama-completion /app
149
150ENTRYPOINT [ "/app/llama-cli" ]
151
152### Target: server
153# Dedicated server image containing only llama-server
154# ==============================================================================
155FROM base AS server
156
157ENV LLAMA_ARG_HOST=0.0.0.0
158
159COPY --from=build /app/full/llama /app/full/llama-server /app
160
161HEALTHCHECK --interval=5m CMD [ "curl", "-f", "http://localhost:8080/health" ]
162
163ENTRYPOINT [ "/app/llama-server" ]