blob: ff61000fd08620cb8589a823bfd537381c469719 (
plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
|
# 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"
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-${TINY_LLAMAS_COMMIT}.gguf
)
"
LICENSE="MIT"
SLOT="0"
CPU_FLAGS_X86=( avx avx2 f16c )
# wwma USE explained here: https://github.com/ggml-org/llama.cpp/blob/master/docs/build.md#hip
IUSE="curl openblas +openmp blis rocm cuda opencl vulkan flexiblas wmma examples"
REQUIRED_USE="
?? (
openblas
blis
flexiblas
)
wmma? (
rocm
)
"
# curl is needed for pulling models from huggingface
# numpy is used by convert_hf_to_gguf.py
CDEPEND="
curl? ( net-misc/curl:= )
openblas? ( sci-libs/openblas:= )
openmp? ( llvm-runtimes/openmp:= )
blis? ( sci-libs/blis:= )
flexiblas? ( sci-libs/flexiblas:= )
rocm? (
>=dev-util/hip-${ROCM_VERSION}:=
>=sci-libs/hipBLAS-${ROCM_VERSION}:=
wmma? (
>=sci-libs/rocWMMA-${ROCM_VERSION}:=
)
)
cuda? ( dev-util/nvidia-cuda-toolkit:= )
"
DEPEND="${CDEPEND}
opencl? ( dev-util/opencl-headers )
vulkan? ( dev-util/vulkan-headers )
"
RDEPEND="${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
}
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)
-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"
}
|
