Merge pull request #1041 from roboflow/fix/reduce-gpu-image-size

Fix the multistage build for smaller image

docker/dockerfiles/Dockerfile.onnx.gpu

@@ -1,4 +1,4 @@
-FROM nvcr.io/nvidia/cuda:11.8.0-cudnn8-devel-ubuntu22.04 as base
+FROM nvcr.io/nvidia/cuda:11.8.0-cudnn8-devel-ubuntu22.04 as builder
 
 WORKDIR /app
 
@@ -49,13 +49,32 @@ RUN python3 -m pip install \
     && rm -rf ~/.cache/pip
 
 # Install setup.py requirements for flash_attn
-RUN python3 -m pip install packaging==24.1  && rm -rf ~/.cache/pip
+RUN python3 -m pip install packaging==24.1 && rm -rf ~/.cache/pip
 
 # Install flash_attn required for Paligemma and Florence2
 RUN python3 -m pip install -r requirements.pali.flash_attn.txt --no-dependencies --no-build-isolation && rm -rf ~/.cache/pip
 
-FROM scratch
-COPY --from=base / /
+# Start runtime stage
+FROM nvcr.io/nvidia/cuda:11.8.0-cudnn8-runtime-ubuntu22.04 as runtime
+
+WORKDIR /app
+
+# Copy Python and installed packages from builder
+COPY --from=builder /usr/local/lib/python3.10 /usr/local/lib/python3.10
+COPY --from=builder /usr/local/bin /usr/local/bin
+
+# Install runtime dependencies
+RUN rm -rf /var/lib/apt/lists/* && apt-get clean && apt-get update -y && DEBIAN_FRONTEND=noninteractive apt-get install -y \
+    libxext6 \
+    libopencv-dev \
+    uvicorn \
+    python3-pip \
+    git \
+    libgdal-dev \
+    wget \
+    rustc \
+    cargo \
+    && rm -rf /var/lib/apt/lists/*
 
 WORKDIR /build
 COPY . .
@@ -86,4 +105,4 @@ ENV ENABLE_WORKFLOWS_PROFILING=True
 ENV ENABLE_PROMETHEUS=True
 ENV STREAM_API_PRELOADED_PROCESSES=2
 
-ENTRYPOINT uvicorn gpu_http:app --workers $NUM_WORKERS --host $HOST --port $PORT
+ENTRYPOINT uvicorn gpu_http:app --workers $NUM_WORKERS --host $HOST --port $PORT

inference/core/version.py

@@ -1,4 +1,4 @@
-__version__ = "0.39.0rc2"
+__version__ = "0.39.0"
 
 
 if __name__ == "__main__":