fix: split nnunetv2 install to preserve CUDA PyTorch in Docker

- Install all deps except nnunetv2 first (keeps base image CUDA torch)
- Install nnunetv2 deps separately, then nnunetv2 with --no-deps
- Add system dev packages for C compilation (SimpleITK, blosc2, etc.)
- Add build verification step to fail loudly if nnunetv2 missing

Dockerfile

@@ -5,11 +5,18 @@
 
 FROM pytorch/pytorch:2.1.0-cuda11.8-cudnn8-runtime
 
-# Install system dependencies
+# Install system dependencies required for nnunetv2 and its transitive deps
+# (SimpleITK, blosc2, imagecodecs need C compilation toolchains)
 RUN apt-get update && apt-get install -y --no-install-recommends \
     build-essential \
     curl \
     git \
+    pkg-config \
+    libjpeg-dev \
+    libpng-dev \
+    libtiff-dev \
+    libopenblas-dev \
+    graphviz \
     && rm -rf /var/lib/apt/lists/*
 
 # Create non-root user (HF Spaces requirement)
@@ -20,10 +27,27 @@ ENV HOME=/home/user \
 
 WORKDIR /home/user/app
 
-# Copy requirements and install Python dependencies
+# Copy requirements first for Docker cache
 COPY --chown=user:user requirements.txt .
+
+# Step 1: Install all deps EXCEPT nnunetv2 (keep base image's CUDA torch)
 RUN pip install --no-cache-dir --upgrade pip && \
-    pip install --no-cache-dir -r requirements.txt
+    grep -v nnunetv2 requirements.txt | pip install --no-cache-dir -r /dev/stdin
+
+# Step 2: Install nnunetv2's non-torch dependencies, then nnunetv2 itself
+#         with --no-deps so it does NOT overwrite the CUDA PyTorch
+RUN pip install --no-cache-dir \
+    acvl-utils batchgenerators batchgeneratorsv2 blosc2 \
+    dynamic-network-architectures graphviz imagecodecs \
+    pandas requests scikit-image scikit-learn seaborn \
+    SimpleITK tifffile tqdm yacs && \
+    pip install --no-cache-dir --no-deps nnunetv2>=2.2
+
+# Verify critical packages installed (fail build early if not)
+RUN python -c "import nnunetv2; print('nnunetv2 OK')" && \
+    python -c "import torch; print('torch', torch.__version__); assert 'cu' in torch.__version__ or torch.cuda.is_available(), 'CUDA torch missing!'" && \
+    python -c "import fastapi; print('fastapi OK')" && \
+    python -c "import monai; print('monai OK')"
 
 # Copy application code
 COPY --chown=user:user . .