Create prebuilt.sh

prebuilt.sh

@@ -0,0 +1,13 @@
+#!/bin/bash
+set -e
+
+echo "=== Building llama-cpp-python with CUDA for ZeroGPU ==="
+export CMAKE_ARGS="-DGGML_CUDA=on"
+export FORCE_CMAKE=1
+
+# The Gradio base image has CUDA 12.x installed.
+# Build llama-cpp-python from source so it links against the actual GPU toolkit.
+pip install --no-cache-dir llama-cpp-python
+pip install --no-cache-dir python-docx pdfplumber chardet
+
+echo "=== Build complete ==="