sci-misc/llama-cpp: Add webui and 0_pre9219

sci-misc/llama-cpp/Manifest

@@ -6,3 +6,8 @@ DIST llama-cpp-0_pre7924.tar.gz 28899921 BLAKE2B b89c8c170d1c2d52390dfef35a4e085
 DIST llama-cpp-0_pre8198.tar.gz 29091642 BLAKE2B b5d02310a7e71fec88e29094aa7936cb520530fb9f81cef58f87c97948e6116734f9a19da5ad8daba1027f8e26308e5fa1f3e91e33a49539929f6e3475b08429 SHA512 bc8aef7cb29a32e0acf90d50cbf426c1a42d99659747203449ebf85f0c9c2fed98fd23b7e50b963dd537d3d6edecf17936d0cf7ae04266df4723e2d4a5c7ecfa
 DIST llama-cpp-0_pre8628.tar.gz 29611654 BLAKE2B 5efeb855b5833ca7a0ba552a142e9fac08773e0f0f31d9d3188c03840d91c91aa22edaf310a6472c9077cd4fb2a48c82371887374fb6d417dd022a18c51ff9e2 SHA512 7b86c37b3f926a0983fe6648629019e9f32001dbf474de9278d94d5e1a9e4c37dfaa9832557555724941c32c437efcbf2e91b0b0ad558936657e14ec5e52f8f0
 DIST llama-cpp-0_pre8838.tar.gz 33811834 BLAKE2B 1223a15237186dc8e11c454d76adfd875a9d6b3ee56c046fc0cf63415c7ebbb4b0dc2da49b7e697ecd9908dad6adcb4f2a5bd4665b7118d2c2508e3b215c1f39 SHA512 d345a2645dcad2564f94efdef45a7681b85f4079d30e68297b42a529ccfe4f50429067dce8685791c914cd59d00843c8e5d02c6ef7ce2d97bb990d6c896f5e9f
+DIST llama-cpp-0_pre9219-uiasset-bundle.css 505965 BLAKE2B cb09d237e482b0b9474b27adf89a2129a6543c9c8b47beb7b5e77d3f4632f866cb0996e19bca0938394bb3a16d63dca716788dd24c3f7389d4545425e1441330 SHA512 8ddda7222c376dff3a2930452d0e86cda39b4047ee2aeb094da5ea98d83ec3bc8a490b8b0e110560bd5802454d965d403a7b6f2c07aa4a02b21c6504aaf1ded2
+DIST llama-cpp-0_pre9219-uiasset-bundle.js 5279222 BLAKE2B b4f31e74787ffb233b0e6772c05202eacaf79af2faf07a24a199f92e5d9eb0e786e94bd8c23e346ec7f6dcb0fb8a594d5de144a3c2079ced05397466d7f256ee SHA512 076d09f41cbbfc1a8fed044802bf4b6b47d1c0b26abecf5958845ab3e99db1aa5d7d5096bfdb1c34a98a5be3ea0cd974c3af3aa13226cf0659b68b08ad86edca
+DIST llama-cpp-0_pre9219-uiasset-index.html 6917 BLAKE2B 4c81aa40f5ab4618e9b1f14d1ac2eaf062d7fd4b11c309c9027432b5e0e9b4bf92dc6555b1a5ca81359beac884ea39c1cd074170e2be4407315fb170ca1c35db SHA512 d48aae733cb64b2fa2893e64fcea68da284e13923572efb174619f85c507825def0757093e8eb54f7bd7cdcb31b63e5a6854daef6e6e25ee0eb7284bb8cc8432
+DIST llama-cpp-0_pre9219-uiasset-loading.html 268 BLAKE2B 4b8339b8b3a52bb151f3ab9eabc4d3187aeea312da255d4f55a80839f8145258ee1cf9ea594abe5aa2b49731fe55716870312095fd6a10df42b6cfa4f120ac70 SHA512 d045fac9f2cf87aa5756f27ef0760df494399de6df40ca0a06387685eb69150053a1204b8b17d2d8a8166279081c01f7fbbba10645510309e9debe18cdc20f0c
+DIST llama-cpp-0_pre9219.tar.gz 33881298 BLAKE2B 27f66cea2798f151bfc9a648fa4197e65dc8e0969e7793416b10bce6659f46e54c8d80e4d951f3ea4081df23909bd9616740cc506401ef32f4e8ccf1a21a26b8 SHA512 e8ec0330802533f6c60c42eca67d3db99d6310d9b3b37c5b7b05cb9dd451c3372e2eb25d9c655e3e396e1f472042401bf3f1b431c24995b32b8c320b4aa3297a

sci-misc/llama-cpp/llama-cpp-0_pre9219.ebuild

@@ -0,0 +1,187 @@
+# Copyright 2026 Gentoo Authors
+# Distributed under the terms of the GNU General Public License v2
+
+EAPI=8
+
+ROCM_VERSION="6.3"
+
+inherit cmake cuda rocm linux-info
+
+TINY_LLAMAS_COMMIT="99dd1a73db5a37100bd4ae633f4cfce6560e1567"
+
+DESCRIPTION="Port of Facebook's LLaMA model in C/C++"
+HOMEPAGE="https://github.com/ggml-org/llama.cpp"
+
+UI_ASSETS=( index.html bundle.js bundle.css loading.html )
+UI_ASSET_BUCKET="https://huggingface.co/buckets/ggml-org/llama-ui"
+
+if [[ ${PV} == *9999* ]]; then
+	inherit git-r3
+	EGIT_REPO_URI="https://github.com/ggml-org/llama.cpp.git"
+else
+	MY_PV="b${PV#0_pre}"
+	SRC_URI="https://github.com/ggml-org/llama.cpp/archive/refs/tags/${MY_PV}.tar.gz -> ${P}.tar.gz"
+	for asset in "${UI_ASSETS[@]}"; do
+		SRC_URI+="
+			webui? ( ${UI_ASSET_BUCKET}/resolve/${MY_PV}/${asset}?download=true -> ${P}-uiasset-${asset} )
+		"
+	done
+	S="${WORKDIR}/llama.cpp-${MY_PV}"
+	KEYWORDS="~amd64"
+fi
+
+SRC_URI+="
+	examples? (
+		https://huggingface.co/ggml-org/tiny-llamas/resolve/${TINY_LLAMAS_COMMIT}/stories15M-q4_0.gguf
+			-> ggml-org_models_tinyllamas_stories15M-q4_0-${TINY_LLAMAS_COMMIT}.gguf
+	)
+"
+
+LICENSE="MIT"
+SLOT="0"
+CPU_FLAGS_X86=( avx avx2 f16c )
+
+# wwma USE explained here: https://github.com/ggml-org/llama.cpp/blob/master/docs/build.md#hip
+IUSE="curl openblas +openmp blis rocm cuda opencl openssl vulkan flexiblas wmma examples webui"
+
+REQUIRED_USE="
+	?? (
+		openblas
+		blis
+		flexiblas
+	)
+	wmma? (
+		rocm
+	)
+"
+if [[ ${PV} == *9999* ]]; then
+	# Webui assets are only published for tagged releases and
+	# building them locally would require fetching npm dependencies
+	REQUIRED_USE+="!webui"
+fi
+
+# curl is needed for pulling models from huggingface
+# numpy is used by convert_hf_to_gguf.py
+CDEPEND="
+	curl? ( net-misc/curl:= )
+	openblas? ( sci-libs/openblas:= )
+	openmp? ( llvm-runtimes/openmp:= )
+	blis? ( sci-libs/blis:= )
+	flexiblas? ( sci-libs/flexiblas:= )
+	rocm? (
+		>=dev-util/hip-${ROCM_VERSION}:=
+		>=sci-libs/hipBLAS-${ROCM_VERSION}:=
+		wmma? (
+			>=sci-libs/rocWMMA-${ROCM_VERSION}:=
+		)
+	)
+	cuda? ( dev-util/nvidia-cuda-toolkit:= )
+	openssl? ( dev-libs/openssl:= )
+"
+DEPEND="${CDEPEND}
+	opencl? ( dev-util/opencl-headers )
+	vulkan? (
+		dev-util/spirv-headers
+		dev-util/vulkan-headers
+	)
+"
+RDEPEND="${CDEPEND}
+	dev-python/numpy
+	opencl? ( dev-libs/opencl-icd-loader )
+	vulkan? ( media-libs/vulkan-loader )
+"
+BDEPEND="media-libs/shaderc"
+
+pkg_setup() {
+	if use rocm; then
+		linux-info_pkg_setup
+		if linux-info_get_any_version && linux_config_exists; then
+			if ! linux_chkconfig_present HSA_AMD_SVM; then
+				ewarn "To use ROCm/HIP, you need to have HSA_AMD_SVM option enabled in your kernel."
+			fi
+		fi
+	fi
+}
+
+src_prepare() {
+	use cuda && cuda_src_prepare
+	cmake_src_prepare
+	if use examples; then
+		mkdir -p "${BUILD_DIR}/tinyllamas" || die
+		cp "${DISTDIR}/ggml-org_models_tinyllamas_stories15M-q4_0-${TINY_LLAMAS_COMMIT}.gguf" \
+			"${BUILD_DIR}/tinyllamas/stories15M-q4_0.gguf" || die
+	fi
+	if use webui; then
+		mkdir -p "${S}/build/tools/ui/dist" || die
+		for asset in "${UI_ASSETS[@]}"; do
+			ln -s "${DISTDIR}/${P}-uiasset-${asset}" "${S}/build/tools/ui/dist/${asset}" || die
+		done
+	fi
+}
+
+src_configure() {
+	local mycmakeargs=(
+		-DLLAMA_BUILD_WEBUI=$(usex webui)
+		-DLLAMA_BUILD_TESTS=OFF
+		-DLLAMA_BUILD_EXAMPLES=$(usex examples)
+		-DLLAMA_BUILD_SERVER=ON
+		-DCMAKE_SKIP_BUILD_RPATH=ON
+		-DGGML_NATIVE=0	# don't set march
+		-DGGML_RPC=ON
+		-DLLAMA_CURL=$(usex curl)
+		-DLLAMA_OPENSSL=$(usex openssl)
+		-DBUILD_NUMBER="1"
+		-DGENTOO_REMOVE_CMAKE_BLAS_HACK=ON
+		-DGGML_CUDA=$(usex cuda)
+		-DGGML_OPENCL=$(usex opencl)
+		-DGGML_OPENMP=$(usex openmp)
+		-DGGML_VULKAN=$(usex vulkan)
+
+		# avoid clashing with whisper.cpp
+		-DCMAKE_INSTALL_LIBDIR="${EPREFIX}/usr/$(get_libdir)/llama.cpp"
+		-DCMAKE_INSTALL_RPATH="${EPREFIX}/usr/$(get_libdir)/llama.cpp"
+	)
+
+	if use openblas ; then
+		mycmakeargs+=(
+			-DGGML_BLAS=ON -DGGML_BLAS_VENDOR=OpenBLAS
+		)
+	fi
+
+	if use blis ; then
+		mycmakeargs+=(
+			-DGGML_BLAS=ON -DGGML_BLAS_VENDOR=FLAME
+		)
+	fi
+
+	if use flexiblas; then
+		mycmakeargs+=(
+			-DGGML_BLAS=ON -DGGML_BLAS_VENDOR=FlexiBLAS
+		)
+	fi
+
+	if use cuda; then
+		local -x CUDAHOSTCXX="$(cuda_gccdir)"
+		# tries to recreate dev symlinks
+		cuda_add_sandbox
+		addpredict "/dev/char/"
+	fi
+
+	if use rocm; then
+		rocm_use_hipcc
+		mycmakeargs+=(
+			-DGGML_HIP=ON -DAMDGPU_TARGETS=$(get_amdgpu_flags)
+			-DGGML_HIP_ROCWMMA_FATTN=$(usex wmma)
+		)
+	fi
+
+	cmake_src_configure
+}
+
+src_install() {
+	cmake_src_install
+	dobin "${BUILD_DIR}/bin/rpc-server"
+
+	# avoid clashing with whisper.cpp
+	rm -rf "${ED}/usr/include"
+}

sci-misc/llama-cpp/metadata.xml

@@ -13,6 +13,7 @@
 		<flag name="openblas">Build an OpenBLAS backend</flag>
 		<flag name="opencl">Build an OpenCL backend, so far only works on Adreno and Intel GPUs</flag>
 		<flag name="openssl">Use openssl to support HTTPS</flag>
+		<flag name="webui">Build llama-server with the webui (not supported in live ebuilds)</flag>
 	</use>
 	<maintainer type="person">
 		<email>zl29ah@gmail.com</email>