chore: add build_llama_server.sh (local cross-compile script, unused)

modal_deploy/build_llama_server.sh

@@ -0,0 +1,43 @@
+#!/bin/bash
+# Build llama-server locally for linux/amd64 (Modal's target)
+# Uses Docker to cross-compile, then uploads binary to Modal Volume
+
+set -e
+
+SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd)"
+PROJECT_DIR="$(dirname "$SCRIPT_DIR")"
+LLAMACPP_DIR="$SCRIPT_DIR/llamacpp_omni"
+OUTPUT_DIR="$SCRIPT_DIR/build_output"
+
+echo "=== PregoPal: Cross-build llama-server for Modal ==="
+echo "Source: $LLAMACPP_DIR"
+echo "Output: $OUTPUT_DIR"
+
+# Clean output
+rm -rf "$OUTPUT_DIR"
+mkdir -p "$OUTPUT_DIR"
+
+# Build with docker using linux/amd64 (Modal's arch)
+docker run --rm \
+  -v "$LLAMACPP_DIR:/llama.cpp-omni" \
+  -v "$OUTPUT_DIR:/output" \
+  -w /llama.cpp-omni \
+  --platform linux/amd64 \
+  nvidia/cuda:12.4.1-devel-ubuntu22.04 \
+  bash -c "
+    apt-get update && apt-get install -y cmake build-essential libcurl4-openssl-dev pkg-config &&
+    cmake -B build \
+      -DGGML_CUDA=ON \
+      -DLLAMA_BUILD_SERVER=ON \
+      -DLLAMA_BUILD_TESTS=OFF \
+      -DLLAMA_BUILD_EXAMPLES=OFF \
+      -DLLAMA_CUDA_FORCE_MMQ=ON \
+      -DCMAKE_CUDA_ARCHITECTURES='75;89' \
+      -DCMAKE_BUILD_TYPE=Release &&
+    cmake --build build --config Release -j \$(nproc) --target llama-server &&
+    cp build/bin/llama-server /output/
+  "
+
+echo "=== Build complete! Binary: $OUTPUT_DIR/llama-server ==="
+ls -lh "$OUTPUT_DIR/llama-server"
+file "$OUTPUT_DIR/llama-server"