Spaces:
Running
Running
Create Dockerfile
Browse files- Dockerfile +37 -0
Dockerfile
ADDED
|
@@ -0,0 +1,37 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
FROM ubuntu:22.04
|
| 2 |
+
|
| 3 |
+
ENV DEBIAN_FRONTEND=noninteractive
|
| 4 |
+
WORKDIR /app
|
| 5 |
+
|
| 6 |
+
# Install dependencies
|
| 7 |
+
RUN apt update && apt install -y \
|
| 8 |
+
git \
|
| 9 |
+
cmake \
|
| 10 |
+
build-essential \
|
| 11 |
+
curl \
|
| 12 |
+
ca-certificates \
|
| 13 |
+
&& rm -rf /var/lib/apt/lists/*
|
| 14 |
+
|
| 15 |
+
# Clone llama.cpp
|
| 16 |
+
RUN git clone https://github.com/ggerganov/llama.cpp
|
| 17 |
+
|
| 18 |
+
# Build llama-server
|
| 19 |
+
RUN cmake -S llama.cpp -B llama.cpp/build \
|
| 20 |
+
-DLLAMA_BUILD_SERVER=ON \
|
| 21 |
+
-DCMAKE_BUILD_TYPE=Release \
|
| 22 |
+
&& cmake --build llama.cpp/build --target llama-server -j$(nproc)
|
| 23 |
+
|
| 24 |
+
# Create model directory
|
| 25 |
+
RUN mkdir -p /models
|
| 26 |
+
|
| 27 |
+
# Download a GGUF model (example: Qwen 0.5B)
|
| 28 |
+
RUN curl -L -o /models/model.gguf \
|
| 29 |
+
https://huggingface.co/Qwen/Qwen2.5-0.5B-Instruct-GGUF/resolve/main/qwen2.5-0.5b-instruct-q4_k_m.gguf
|
| 30 |
+
|
| 31 |
+
# Copy start script
|
| 32 |
+
COPY start.sh /start.sh
|
| 33 |
+
RUN chmod +x /start.sh
|
| 34 |
+
|
| 35 |
+
EXPOSE 7860
|
| 36 |
+
|
| 37 |
+
CMD ["/start.sh"]
|