mirror of
https://github.com/gentoo-mirror/guru.git
synced 2026-04-07 03:10:15 -04:00
sci-misc/llama-cpp: add 0_pre8149, drop 0_pre6710
for Qwen3.5 support Signed-off-by: Sergey Alirzaev <l29ah@riseup.net>
This commit is contained in:
Sergey Alirzaev
@@ -1,8 +1,8 @@
|
||||
DIST ggml-org_models_tinyllamas_stories15M-q4_0-99dd1a73db5a37100bd4ae633f4cfce6560e1567.gguf 19077344 BLAKE2B 16e65adf9785e3091c51f1de59e5580f93fb47f79961513aeb3dbb8a0f5930f7120f0304f0f293a006170805e2b70ee1fcff0496b63356323d32c2caa55be8a8 SHA512 f9944886089958e0d97b1906cfd45020e0821c65429346e76fae29136c634ae5d039dffbae5933a95b0674f4acd87b656feb9f9e1b16dd434c5c9b5886f4f617
|
||||
DIST llama-cpp-0_pre6710.tar.gz 25894417 BLAKE2B 147f30d76fd49bf18fa0ab9e3e75d0ad337dcd87a73f1dbce43f180488ea06b40b1a2a93b4686a88b5a442dd4dd6a8e45bf848ceb549bdc0ad0078427336c56e SHA512 75c5918713256cb11f704b94d6e249a9f3ac2dde1107a6f4506134ba9c772e1c42d991915b571887207003f4b0679a183cd0787ffd742a08d2283fdfb86695eb
|
||||
DIST llama-cpp-0_pre6980.tar.gz 26431911 BLAKE2B b7d7c0dcdabde01acb816e73bc344564823dd1fc498fb98bf3c611b2d7a964af4d94f7cad533fe675a30685d510829160e392ab0f3bd16f4757a2f3446b8e3ca SHA512 33e63336ad7c0fc653acd409d9314ce3fc3755ed1c03b4806c647b7c80d91b3c883aec6633334555c3855a24276d4975a54c96af91df8d2f818d4dd1dbcbabfb
|
||||
DIST llama-cpp-0_pre7276.tar.gz 27765814 BLAKE2B d0553ab1dd29c9d93a18c6217aab4553faf09e385a94b90732a537bbcf9bded54d5cda28553543e2c0cc71b6a157bfb80a48405f3f8281c51525757967b33e16 SHA512 3035fe53fea2ca3b0f35e479f4eaec75e38a2ea670600445776cd6fa696fc83ca19eb6dd7cd2ab1da69e78293c62318b5182e6e5b3423ae6c1f00854c5132a4c
|
||||
DIST llama-cpp-0_pre7611.tar.gz 28622786 BLAKE2B 3c345645c9bcf07d8a513b9e883619b31b5254581f73429d638403758429fd2dfc5f78a22d538e8d88eb6c1be74bf805481af697480727ed750492ddec5c37fe SHA512 c6c4780d7e68adfc385b57c6f7530423f8205bfa283572b0d414d55e143c03307e98676e41ad527c37d7837f831f8ff24be0f7bf59e366ea82f3802cdc946821
|
||||
DIST llama-cpp-0_pre7836.tar.gz 28813563 BLAKE2B c843c0199b528114c23c58536552e2a7f581fda93364353e8aa38fda99ddb1942c421ad00be4daf86b49a2c311ef89bb6908b8900ca1ffb64df338cbfa11c354 SHA512 161176107de175d9b6fdc1ebfa9fe0cf6fd6968245730f7df7f17c6a179ef111685d0fed5a45031825dc4b07db270887a4307061f9918e375a1c65f261062d9d
|
||||
DIST llama-cpp-0_pre7924.tar.gz 28899921 BLAKE2B b89c8c170d1c2d52390dfef35a4e0857b4a3aa174077f2fdfa0fbbe1a254c5e6ae9b976bde9f52e42ed3932403aec728b9c415ca44ccd22c061e7e8a1e481526 SHA512 689c73215e795a53be1d0e6a639983aded1ed7bdeb3a261556de3db2cfdc765a7daf4aa142433f5e99a255dc123035a929b57868d3941ddda400a3c2b1db31cb
|
||||
DIST llama-cpp-0_pre8100.tar.gz 29031553 BLAKE2B daa0f439391dd9b0b7804b893d060f11c94600dc4347c0f4d86ede376ef951f725d4811c07b729e6e7bb06cab5e7e6aacc5c3a99b8e6e13ff3e7ed2ae5eae288 SHA512 2582732ee28839d940f5feb66640123070eb0a57f0b6c434fcdca85c84ccabccff430b01c5d361b1b936305b8e9fce9886c6038c44d62e084e6ca9a241533821
|
||||
DIST llama-cpp-0_pre8149.tar.gz 29070799 BLAKE2B 6ce513eccbd9bc90b2fb9d054f3d939901dedc2f3036853666180038553e05f1c8d164c373e5be4fe2c2c8c0f91449ec67c069bc4aee316fc3a878f3385f92d7 SHA512 ced93176831563f066be16b3f36b3368ec72ee51be4e4403e15fc78f72daac26f724947a2a2e04bf2211000a6e790a5530b6de48333abb820f99aa2f580a1ef5
|
||||
|
||||
@@ -1,4 +1,4 @@
|
||||
# Copyright 2025 Gentoo Authors
|
||||
# Copyright 2026 Gentoo Authors
|
||||
# Distributed under the terms of the GNU General Public License v2
|
||||
|
||||
EAPI=8
|
||||
@@ -7,24 +7,45 @@ ROCM_VERSION="6.3"
|
||||
|
||||
inherit cmake cuda rocm linux-info
|
||||
|
||||
if [[ "${PV}" != "9999" ]]; then
|
||||
KEYWORDS="~amd64"
|
||||
MY_PV="b${PV#0_pre}"
|
||||
S="${WORKDIR}/llama.cpp-${MY_PV}"
|
||||
SRC_URI="https://github.com/ggml-org/llama.cpp/archive/refs/tags/${MY_PV}.tar.gz -> ${P}.tar.gz"
|
||||
else
|
||||
inherit git-r3
|
||||
EGIT_REPO_URI="https://github.com/ggml-org/llama.cpp.git"
|
||||
fi
|
||||
TINY_LLAMAS_COMMIT="99dd1a73db5a37100bd4ae633f4cfce6560e1567"
|
||||
|
||||
DESCRIPTION="Port of Facebook's LLaMA model in C/C++"
|
||||
HOMEPAGE="https://github.com/ggml-org/llama.cpp"
|
||||
|
||||
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 -> ${P}.tar.gz"
|
||||
S="${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-${TINY_LLAMAS_COMMIT}.gguf
|
||||
)
|
||||
"
|
||||
|
||||
LICENSE="MIT"
|
||||
SLOT="0"
|
||||
CPU_FLAGS_X86=( avx avx2 f16c )
|
||||
IUSE="curl openblas +openmp blis hip cuda opencl vulkan"
|
||||
REQUIRED_USE="?? ( openblas blis )"
|
||||
|
||||
# wwma USE explained here: https://github.com/ggml-org/llama.cpp/blob/master/docs/build.md#hip
|
||||
IUSE="curl openblas +openmp blis rocm cuda opencl vulkan flexiblas wmma examples"
|
||||
|
||||
REQUIRED_USE="
|
||||
?? (
|
||||
openblas
|
||||
blis
|
||||
flexiblas
|
||||
)
|
||||
wmma? (
|
||||
rocm
|
||||
)
|
||||
"
|
||||
|
||||
# curl is needed for pulling models from huggingface
|
||||
# numpy is used by convert_hf_to_gguf.py
|
||||
@@ -33,8 +54,13 @@ CDEPEND="
|
||||
openblas? ( sci-libs/openblas:= )
|
||||
openmp? ( llvm-runtimes/openmp:= )
|
||||
blis? ( sci-libs/blis:= )
|
||||
hip? ( >=dev-util/hip-6.3:=
|
||||
>=sci-libs/hipBLAS-6.3:=
|
||||
flexiblas? ( sci-libs/flexiblas:= )
|
||||
rocm? (
|
||||
>=dev-util/hip-${ROCM_VERSION}:=
|
||||
>=sci-libs/hipBLAS-${ROCM_VERSION}:=
|
||||
wmma? (
|
||||
>=sci-libs/rocWMMA-${ROCM_VERSION}:=
|
||||
)
|
||||
)
|
||||
cuda? ( dev-util/nvidia-cuda-toolkit:= )
|
||||
"
|
||||
@@ -50,37 +76,41 @@ RDEPEND="${CDEPEND}
|
||||
BDEPEND="media-libs/shaderc"
|
||||
|
||||
pkg_setup() {
|
||||
if use hip; then
|
||||
if use rocm; then
|
||||
linux-info_pkg_setup
|
||||
if linux-info_get_any_version && linux_config_exists; then
|
||||
if ! linux_chkconfig_present HSA_AMD_SVM; then
|
||||
ewarn "To use ROCm/HIP, you need to have HSA_AMD_SVM option enabled in your kernel."
|
||||
fi
|
||||
fi
|
||||
|
||||
fi
|
||||
}
|
||||
|
||||
src_prepare() {
|
||||
use cuda && cuda_src_prepare
|
||||
|
||||
cmake_src_prepare
|
||||
if use examples; then
|
||||
mkdir -p "${BUILD_DIR}/tinyllamas" || die
|
||||
cp "${DISTDIR}/ggml-org_models_tinyllamas_stories15M-q4_0-${TINY_LLAMAS_COMMIT}.gguf" \
|
||||
"${BUILD_DIR}/tinyllamas/stories15M-q4_0.gguf" || die
|
||||
fi
|
||||
}
|
||||
|
||||
src_configure() {
|
||||
local mycmakeargs=(
|
||||
-DLLAMA_BUILD_TESTS=OFF
|
||||
-DLLAMA_BUILD_EXAMPLES=$(usex examples)
|
||||
-DLLAMA_BUILD_SERVER=ON
|
||||
-DCMAKE_SKIP_BUILD_RPATH=ON
|
||||
-DGGML_NATIVE=0 # don't set march
|
||||
-DGGML_RPC=ON
|
||||
-DLLAMA_CURL=$(usex curl ON OFF)
|
||||
-DLLAMA_CURL=$(usex curl)
|
||||
-DBUILD_NUMBER="1"
|
||||
-DGENTOO_REMOVE_CMAKE_BLAS_HACK=ON
|
||||
-DGGML_CUDA=$(usex cuda ON OFF)
|
||||
-DGGML_OPENCL=$(usex opencl ON OFF)
|
||||
-DGGML_OPENMP=$(usex openmp ON OFF)
|
||||
-DGGML_VULKAN=$(usex vulkan ON OFF)
|
||||
-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"
|
||||
@@ -99,6 +129,12 @@ src_configure() {
|
||||
)
|
||||
fi
|
||||
|
||||
if use flexiblas; then
|
||||
mycmakeargs+=(
|
||||
-DGGML_BLAS=ON -DGGML_BLAS_VENDOR=FlexiBLAS
|
||||
)
|
||||
fi
|
||||
|
||||
if use cuda; then
|
||||
local -x CUDAHOSTCXX="$(cuda_gccdir)"
|
||||
# tries to recreate dev symlinks
|
||||
@@ -106,10 +142,11 @@ src_configure() {
|
||||
addpredict "/dev/char/"
|
||||
fi
|
||||
|
||||
if use hip; then
|
||||
if use rocm; then
|
||||
rocm_use_hipcc
|
||||
mycmakeargs+=(
|
||||
-DGGML_HIP=ON -DAMDGPU_TARGETS=$(get_amdgpu_flags)
|
||||
-DGGML_HIP_ROCWMMA_FATTN=$(usex wmma)
|
||||
)
|
||||
fi
|
||||
|
||||
Reference in New Issue
Block a user