From 6cb0d11453ee21a65d3df548b67896f2e02c74b1 Mon Sep 17 00:00:00 2001
From: tikikun <daogiatuank54@gmail.com>
Date: Wed, 17 Jan 2024 15:41:28 +0700
Subject: [PATCH] feat: add example GPU version of nitro on docker image

---
 examples/example-docker/cuda.Dockerfile | 30 +++++++++++++++++++++++++
 1 file changed, 30 insertions(+)
 create mode 100644 examples/example-docker/cuda.Dockerfile

diff --git a/examples/example-docker/cuda.Dockerfile b/examples/example-docker/cuda.Dockerfile
new file mode 100644
index 000000000..184d8fc5c
--- /dev/null
+++ b/examples/example-docker/cuda.Dockerfile
@@ -0,0 +1,30 @@
+# Use Debian stable slim as the base image
+FROM nvidia/cuda:12.3.1-devel-ubuntu22.04 
+
+# Set working directory
+WORKDIR /app
+
+# Install required packages
+RUN apt-get update && \
+    apt-get install -y git cmake numactl uuid-dev && \
+    git clone --recurse https://github.com/janhq/nitro nitro && \
+    cd nitro && \
+    ./install_deps.sh && \
+    mkdir build && \
+    cd build && \
+    cmake .. -DDEBUG=ON -DLLAMA_CUBLAS=ON -DLLAMA_CUDA_F16=ON -DLLAMA_CUDA_DMMV_X=64 -DLLAMA_CUDA_MMV_Y=32 && \
+    cmake --build . --config Release -j $(nproc) && \
+    apt-get remove --purge -y git cmake && \
+    apt-get autoremove -y && \
+    apt-get clean && \
+    rm -rf /var/lib/apt/lists/*
+
+# Expose the port
+EXPOSE 3928
+
+# Change the permissions of the nitro binary to make it executable
+RUN chmod +x /app/nitro/build/nitro
+
+# Set the command to run the nitro binary with numactl limiting to cores 0-7
+ENTRYPOINT ["/app/nitro/build/nitro"]
+CMD ["1", "0.0.0.0", "3928"]