openPangu-2.0-Flash support for llama.cpp v0.7.0 (diff + patch.sh)

patch.sh

@@ -0,0 +1,73 @@
+#!/usr/bin/env bash
+# ----------------------------------------------------------------------------
+# openPangu-2.0-Flash support + speed optimizations for llama.cpp v0.7.0
+#
+# Downloads the pinned upstream llama.cpp commit, applies the patch and
+# prints build/run instructions.
+#
+# Usage: ./patch.sh
+# ----------------------------------------------------------------------------
+set -euo pipefail
+
+LLAMA_COMMIT="620d2225cc436ab0b001e3200e590e3b286ef9ab"   # llama.cpp release v0.7.0
+SRC_DIR="llama.cpp-${LLAMA_COMMIT}"
+
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+PATCH_FILE="${SCRIPT_DIR}/llama.cpp-openpangu-v0.7.0.diff"
+
+if [ ! -f "${PATCH_FILE}" ]; then
+    echo "error: ${PATCH_FILE} not found (keep it next to this script)" >&2
+    exit 1
+fi
+
+cd "${SCRIPT_DIR}"
+
+if [ ! -d "${SRC_DIR}" ]; then
+    echo "==> downloading llama.cpp @ ${LLAMA_COMMIT} (v0.7.0)"
+    if command -v curl >/dev/null 2>&1; then
+        curl -L --fail -o llama.tar.gz \
+            "https://codeload.github.com/ggml-org/llama.cpp/tar.gz/${LLAMA_COMMIT}"
+    elif command -v wget >/dev/null 2>&1; then
+        wget -O llama.tar.gz \
+            "https://codeload.github.com/ggml-org/llama.cpp/tar.gz/${LLAMA_COMMIT}"
+    else
+        echo "error: need curl or wget" >&2
+        exit 1
+    fi
+    echo "==> extracting"
+    tar xzf llama.tar.gz
+    rm -f llama.tar.gz
+fi
+
+cd "${SRC_DIR}"
+
+echo "==> applying $(basename "${PATCH_FILE}")"
+if command -v git >/dev/null 2>&1 && git apply --check "${PATCH_FILE}" 2>/dev/null; then
+    git apply "${PATCH_FILE}"
+elif command -v patch >/dev/null 2>&1; then
+    patch -p1 --forward < "${PATCH_FILE}"
+else
+    echo "error: need git or patch to apply the diff" >&2
+    exit 1
+fi
+
+echo
+echo "==> done. build with:"
+echo
+echo "    cd ${SRC_DIR}"
+echo "    cmake -B build -S . -DCMAKE_BUILD_TYPE=Release"
+echo "    cmake --build build -j --target llama-cli llama-server"
+echo
+echo "    # CPU only (no GPU):"
+echo "    cmake -B build -S . -DCMAKE_BUILD_TYPE=Release -DGGML_NATIVE=ON"
+echo "    # CUDA:"
+echo "    cmake -B build -S . -DCMAKE_BUILD_TYPE=Release -DGGML_CUDA=ON"
+echo
+echo "==> run (get the GGUF models at https://huggingface.co/mrexodia/openPangu-2.0-Flash-GGUF):"
+echo
+echo "    ./build/bin/llama-cli -m openPangu-2.0-Flash-base-Q3_K_M.gguf -c 65536 --jinja -t \$(nproc)"
+echo "    ./build/bin/llama-server -m openPangu-2.0-Flash-base-Q3_K_M.gguf -c 65536 --jinja --host 0.0.0.0 --port 8080"
+echo
+echo "    # optional multi-token-prediction self-speculative decoding (needs the split MTP head):"
+echo "    ./build/bin/llama-cli -m openPangu-2.0-Flash-base-Q3_K_M.gguf \\"
+echo "        --spec-draft-model openPangu-2.0-Flash-mtp-Q8_0.gguf --spec-type draft-mtp -c 65536 --jinja"