Dockerfile

FROM ubuntu:22.04

RUN apt-get update && apt-get install -y \
    unzip wget curl python3 python3-pip

# Sao chép binary LLaMA đã build
COPY llama_bin.zip /app/llama_bin.zip
WORKDIR /app

# Giải nén binaries
RUN unzip llama_bin.zip && rm llama_bin.zip && chmod +x bin/*

# Đặt biến môi trường để tìm `libllama.so`
ENV LD_LIBRARY_PATH=/app/bin:$LD_LIBRARY_PATH

# Sao chép model
RUN mkdir -p /models
RUN wget -O /models/qwen2.5-0.5b-instruct-q5_k_m.gguf \
    https://huggingface.co/Qwen/Qwen2.5-0.5B-Instruct-GGUF/resolve/main/qwen2.5-0.5b-instruct-q5_k_m.gguf

# Chạy server
CMD ["./bin/llama-server", "-m", "/models/qwen2.5-0.5b-instruct-q5_k_m.gguf", "-p", "8000"]