From 32f7e6ca56f6ea544441e470124662c8cc436a8c Mon Sep 17 00:00:00 2001
From: txtsd <code@ihavea.quest>
Date: Sun, 27 Oct 2024 01:24:03 +0530
Subject: [PATCH] upgpkg: llama.cpp-hip b3982-1

Signed-off-by: txtsd <code@ihavea.quest>
---
 .SRCINFO          | 37 +++++++++++++++++++++++
 .gitignore        |  6 ++++
 .nvchecker.toml   |  4 +++
 PKGBUILD          | 77 +++++++++++++++++++++++++++++++++++++++++++++++
 llama.cpp.conf    |  1 +
 llama.cpp.service | 13 ++++++++
 6 files changed, 138 insertions(+)
 create mode 100644 .SRCINFO
 create mode 100644 .gitignore
 create mode 100644 .nvchecker.toml
 create mode 100644 PKGBUILD
 create mode 100644 llama.cpp.conf
 create mode 100644 llama.cpp.service

diff --git a/.SRCINFO b/.SRCINFO
new file mode 100644
index 0000000000000..5dcdd283f5166
--- /dev/null
+++ b/.SRCINFO
@@ -0,0 +1,37 @@
+pkgbase = llama.cpp-hip
+	pkgdesc = Port of Facebook's LLaMA model in C/C++ (with AMD ROCm optimizations)
+	pkgver = b3982
+	pkgrel = 1
+	url = https://github.com/ggerganov/llama.cpp
+	arch = x86_64
+	arch = armv7h
+	arch = aarch64
+	license = MIT
+	makedepends = rocm-hip-runtime
+	makedepends = rocm-hip-sdk
+	makedepends = clblast
+	makedepends = cmake
+	makedepends = git
+	depends = curl
+	depends = gcc-libs
+	depends = glibc
+	depends = python
+	depends = python-numpy
+	depends = python-sentencepiece
+	depends = hipblas
+	depends = hip-runtime-amd
+	depends = openmp
+	depends = rocblas
+	provides = llama.cpp
+	conflicts = llama.cpp
+	options = lto
+	source = git+https://github.com/ggerganov/llama.cpp#tag=b3982
+	source = git+https://github.com/nomic-ai/kompute.git
+	source = llama.cpp.conf
+	source = llama.cpp.service
+	sha256sums = 24615514be685a63a7d5ce3d7823e27abc09f67fef97bd3c021d1342d1cd4942
+	sha256sums = SKIP
+	sha256sums = 53fa70cfe40cb8a3ca432590e4f76561df0f129a31b121c9b4b34af0da7c4d87
+	sha256sums = 065f69ccd7ac40d189fae723b58d6de2a24966e9b526e0dbfa3035a4c46a7669
+
+pkgname = llama.cpp-hip
diff --git a/.gitignore b/.gitignore
new file mode 100644
index 0000000000000..5cdbded2a16af
--- /dev/null
+++ b/.gitignore
@@ -0,0 +1,6 @@
+pkg/
+src/
+llama.cpp/
+kompute/
+
+*.tar.*
diff --git a/.nvchecker.toml b/.nvchecker.toml
new file mode 100644
index 0000000000000..93c61f0408c8a
--- /dev/null
+++ b/.nvchecker.toml
@@ -0,0 +1,4 @@
+["llama.cpp-hip"]
+source = "git"
+git = "https://github.com/ggerganov/llama.cpp.git"
+include_regex = "b\\d+"
diff --git a/PKGBUILD b/PKGBUILD
new file mode 100644
index 0000000000000..a5a44664fcc85
--- /dev/null
+++ b/PKGBUILD
@@ -0,0 +1,77 @@
+# Maintainer: txtsd <aur.archlinux@ihavea.quest>
+
+pkgname=llama.cpp-hip
+_pkgname=${pkgname%%-hip}
+pkgver=b3982
+pkgrel=1
+pkgdesc="Port of Facebook's LLaMA model in C/C++ (with AMD ROCm optimizations)"
+arch=('x86_64' 'armv7h' 'aarch64')
+url='https://github.com/ggerganov/llama.cpp'
+license=('MIT')
+depends=(
+  'curl'
+  'gcc-libs'
+  'glibc'
+  'python'
+  'python-numpy'
+  'python-sentencepiece'
+  'hipblas'
+  'hip-runtime-amd'
+  'openmp'
+  'rocblas'
+)
+makedepends=(
+  'rocm-hip-runtime'
+  'rocm-hip-sdk'
+  'clblast'
+  'cmake'
+  'git'
+)
+provides=(${_pkgname})
+conflicts=(${_pkgname})
+options=(lto)
+source=(
+  "git+${url}#tag=${pkgver}"
+  "git+https://github.com/nomic-ai/kompute.git"
+  'llama.cpp.conf'
+  'llama.cpp.service'
+)
+sha256sums=('24615514be685a63a7d5ce3d7823e27abc09f67fef97bd3c021d1342d1cd4942'
+            'SKIP'
+            '53fa70cfe40cb8a3ca432590e4f76561df0f129a31b121c9b4b34af0da7c4d87'
+            '065f69ccd7ac40d189fae723b58d6de2a24966e9b526e0dbfa3035a4c46a7669')
+
+prepare() {
+  cd "${_pkgname}"
+
+  git submodule init
+  git config submodule.kompute.url "${srcdir}/kompute"
+  git -c protocol.file.allow=always submodule update
+}
+
+build() {
+  cd "${_pkgname}"
+  export CC=/opt/rocm/llvm/bin/clang
+  export CXX=/opt/rocm/llvm/bin/clang++
+  cmake -S . -B build \
+    -DCMAKE_INSTALL_PREFIX=/usr \
+    -DBUILD_SHARED_LIBS=ON \
+    -DGGML_ALL_WARNINGS_3RD_PARTY=ON \
+    -DGGML_BLAS=ON \
+    -DGGML_LTO=ON \
+    -DGGML_RPC=ON \
+    -DLLAMA_CURL=ON \
+    -DLLAMA_FATAL_WARNINGS=ON \
+    -DGGML_HIPBLAS=ON
+  cmake --build build --config Release
+}
+
+package() {
+  cd "${_pkgname}"
+
+  DESTDIR="${pkgdir}" cmake --install build
+  install -Dm644 'LICENSE' "${pkgdir}/usr/share/licenses/${pkgname}/LICENSE"
+
+  install -Dm644 "${srcdir}/llama.cpp.conf" "${pkgdir}/etc/conf.d/llama.cpp"
+  install -Dm644 "${srcdir}/llama.cpp.service" "${pkgdir}/usr/lib/systemd/system/llama.cpp.service"
+}
diff --git a/llama.cpp.conf b/llama.cpp.conf
new file mode 100644
index 0000000000000..eab1f2be8d5cf
--- /dev/null
+++ b/llama.cpp.conf
@@ -0,0 +1 @@
+LLAMA_ARGS=""
diff --git a/llama.cpp.service b/llama.cpp.service
new file mode 100644
index 0000000000000..4678d85bd1ef4
--- /dev/null
+++ b/llama.cpp.service
@@ -0,0 +1,13 @@
+[Unit]
+Description=llama.cpp Server
+After=syslog.target network.target local-fs.target remote-fs.target nss-lookup.target
+
+[Service]
+Type=simple
+EnvironmentFile=/etc/conf.d/llama.cpp
+ExecStart=/usr/bin/llama-server $LLAMA_ARGS
+ExecReload=/bin/kill -s HUP $MAINPID
+Restart=never
+
+[Install]
+WantedBy=multi-user.target