Update prebuilt.sh

prebuilt.sh

@@ -1,23 +1,14 @@
 #!/bin/bash
 set -e
-
 echo "==============================================="
-echo " Building llama-cpp-python with CUDA for ZeroGPU"
+echo " Installing llama-cpp-python CUDA for ZeroGPU"
 echo "==============================================="
-
-# Your Space has CUDA 13.0.2 toolkit (visible in build logs).
-# Build llama-cpp-python from source linked against it.
-export CMAKE_ARGS="-DGGML_CUDA=on -DCMAKE_CUDA_ARCHITECTURES=native"
-export FORCE_CMAKE=1
-export GGML_CUDA_FORCE_MMQ=on       # force CUDA for all layers
-export GGML_CUDA_FULL_TENSORS=on    # use CUDA tensors throughout
-
-echo "CMAKE_ARGS=$CMAKE_ARGS"
-echo "FORCE_CMAKE=$FORCE_CMAKE"
-
-pip install --no-cache-dir --no-build-isolation llama-cpp-python
-
-echo ""
+# Use the pre-built CUDA 12.4 wheel — installs in seconds, works on CUDA 13
+pip install --no-cache-dir llama-cpp-python \
+  --extra-index-url https://abetlen.github.io/llama-cpp-python/whl/cu124
+echo "[OK] llama-cpp-python CUDA wheel installed"
+# Extra deps
+pip install --no-cache-dir python-docx pdfplumber chardet
 echo "==============================================="
-echo " llama-cpp-python CUDA build complete!"
+echo " Ready."
 echo "==============================================="