Dockerfile.llama-cpp - master - Watashicuvu/agentic-tools

Dockerfile.llama-cpp
63 строки2 KB
Якуб
mts update
12 апр 2026, 21:19
12 апр 2026, 21:19bff6dcd
# Dockerfile для llama.cpp Server с nomic-embed-text-v1.5 для эмбеддингов
#
# Модель загружается из HuggingFace при старте контейнера.
#
# Использование:
#   docker compose up -d llama-cpp-embed

FROM ghcr.io/ggml-org/llama.cpp:server

ENV LLAMA_NO_METAL=1
ENV LLAMA_PORT=9191
ENV LLAMA_MODEL=nomic-ai/nomic-embed-text-v1.5-GGUF
ENV LLAMA_MODEL_FILE=nomic-embed-text-v1.5.Q8_0.gguf

WORKDIR /models

# Скрипт загрузки и запуска модели
RUN echo '#!/bin/bash\n\
set -e\n\
\n\
MODEL_PATH="/models/${LLAMA_MODEL_FILE}"\n\
\n\
# Скачиваем модель из HuggingFace\n\
if [ ! -f "$MODEL_PATH" ]; then\n\
    echo "📥 Downloading embedding model: ${LLAMA_MODEL}"\n\
    echo "   File: ${LLAMA_MODEL_FILE}"\n\
    echo ""\n\
    \n\
    HF_URL="https://huggingface.co/${LLAMA_MODEL}/resolve/main/${LLAMA_MODEL_FILE}"\n\
    \n\
    if wget -q --show-progress -O "$MODEL_PATH" "$HF_URL" 2>/dev/null; then\n\
        echo "✅ Downloaded successfully"\n\
    else\n\
        echo "❌ ERROR: Failed to download model from $HF_URL"\n\
        echo "Please download manually:"\n\
        echo "  wget -O ${MODEL_PATH} ${HF_URL}"\n\
        exit 1\n\
    fi\n\
else\n\
    echo "✅ Embedding model already exists: $MODEL_PATH"\n\
fi\n\
\n\
echo "File size: $(ls -lh "$MODEL_PATH" | awk "{print \$5}")"\n\
\n\
# Запуск llama.cpp server для embeddings\n\
exec /build/bin/llama-server \\\n\
    --model "$MODEL_PATH" \\\n\
    --host 0.0.0.0 \\\n\
    --port ${LLAMA_PORT} \\\n\
    --ctx-size 2048 \\\n\
    --batch-size 512 \\\n\
    --embeddings \\\n\
    --pooling mean \\\n\
    "$@"\n\
' > /entrypoint.sh && chmod +x /entrypoint.sh

EXPOSE 9191

HEALTHCHECK --interval=30s --timeout=10s --start-period=60s --retries=3 \
    CMD curl -f http://localhost:9191/health || exit 1

ENTRYPOINT ["/entrypoint.sh"]
CMD []