blob: fd5ba7fbc1f7bbb67876cfbb1f13d9c499cf3819 (
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
|
# Maintainer: Smoolak <smoolak [at] gmail.com>
pkgname=python-tensorrt-llm
_pkgname=TensorRT-LLM
pkgver=1.2.0rc4
pkgrel=3
pkgdesc="TensorRT LLM provides users with an easy-to-use Python API to define Large Language Models (LLMs) and supports state-of-the-art optimizations to perform inference efficiently on NVIDIA GPUs"
arch=('x86_64')
url="https://github.com/NVIDIA/TensorRT-LLM"
license=('Apache-2.0')
depends=(
'python'
'python-pytorch-cuda'
'python-numpy'
'python-transformers'
'python-tensorrt'
'python-nvidia-modelopt'
'python-accelerate'
'python-diffusers'
'python-onnx'
'python-pillow'
'python-sentencepiece'
'python-psutil'
'python-pydantic'
'python-fastapi'
'uvicorn'
'python-click'
'python-mpi4py'
'python-pandas'
'python-h5py'
'python-openai'
'python-pyzmq'
'python-peft'
'python-einops'
'python-opencv'
'python-soundfile'
'python-tiktoken'
'python-colored'
'python-lark-parser'
'python-pulp'
'python-prometheus_client'
'python-omegaconf'
'python-wheel'
'python-nvidia-ml-py'
'python-torchvision'
'python-optimum'
'python-datasets'
'python-mpmath'
'python-starlette'
'python-ordered-set'
'python-pypatchelf'
'python-jsonschema'
'python-backoff'
'python-matplotlib'
'python-plotly'
'python-numexpr'
'python-triton'
'python-nvtx'
'cuda'
'cudnn'
'tensorrt'
'nccl'
'rdma-core'
)
makedepends=(
'python-build'
'python-installer'
'python-setuptools'
'cmake>=3.27'
'conan'
'ninja'
'meson'
'gcc'
'git'
'python-cutlass'
)
optdepends=(
'python-evaluate: for evaluation metrics'
'python-blobfile: for blob file operations'
'triton-inference-server: for Triton deployment'
)
source=("${pkgname}-${pkgver}.tar.gz::https://github.com/NVIDIA/${_pkgname}/archive/refs/tags/v${pkgver}.tar.gz"
"glog-0.7-compatibility.patch")
sha256sums=('295dffddbd57f6a6b815969e42d171bd42140c3f2b15f68323fc4cac3e97d39a'
'8227c455479f0013ebf65b2fd7541e19e66b323cdb6e2a65d5f4486622484e0e')
prepare() {
cd "${_pkgname}-${pkgver}"
# Apply glog 0.7.0+ compatibility patch
patch -Np1 -i "${srcdir}/glog-0.7-compatibility.patch"
# Patch build_wheel.py to skip pip install (dependencies handled by pacman/AUR)
# Replace the pip install line for requirements
sed -i '/build_run.*pip install -r.*requirements_file/c\ if os.environ.get("SKIP_PIP_INSTALL") != "1":\n build_run(f'"'"'"{venv_python}" -m pip install -r "{requirements_file}"'"'"')\n else:\n print(f"-- Skipping pip install - using system packages...")' scripts/build_wheel.py
# Replace the pip install line for conan
sed -i '/build_run.*pip install conan/c\ if os.environ.get("SKIP_PIP_INSTALL") != "1":\n build_run(f'"'"'"{venv_python}" -m pip install conan==2.14.0'"'"')\n else:\n print(f"-- Skipping conan pip install - using system conan...")' scripts/build_wheel.py
# Patch CMakeLists.txt to skip CUTLASS library setup (which tries to pip install)
# Comment out the execute_process that calls setup_library.py develop --user
sed -i '/execute_process(/,/RESULT_VARIABLE _CUTLASS_LIBRARY_SUCCESS)/s/^/#/' cpp/tensorrt_llm/kernels/cutlass_kernels/CMakeLists.txt
# Set the success variable to 0 to skip the error check
sed -i '/if(NOT _CUTLASS_LIBRARY_SUCCESS EQUAL 0)/i set(_CUTLASS_LIBRARY_SUCCESS 0)' cpp/tensorrt_llm/kernels/cutlass_kernels/CMakeLists.txt
# Set CUDA_HOME for the build
export CUDA_HOME=/opt/cuda
export CUDACXX=/opt/cuda/bin/nvcc
export PATH="/opt/cuda/bin:${PATH}"
}
build() {
cd "${_pkgname}-${pkgver}"
# Set environment variables for CUDA and TensorRT
export CUDA_HOME=/opt/cuda
export CUDACXX=/opt/cuda/bin/nvcc
export TENSORRT_DIR=/usr
export LD_LIBRARY_PATH=/usr/lib:${LD_LIBRARY_PATH}
export PATH="/opt/cuda/bin:${PATH}"
# Build the C++ libraries first using the build_wheel.py script
# This will use CMake and Conan to build the C++ components
# Use --no-venv to avoid PEP 668 issues with system Python
# Set SKIP_PIP_INSTALL=1 to skip pip install (dependencies handled by pacman)
export SKIP_PIP_INSTALL=1
python scripts/build_wheel.py --clean --build_type=Release --no-venv
}
check() {
cd "${_pkgname}-${pkgver}"
# Basic import test (may fail if CUDA/TensorRT not available during build)
PYTHONPATH="${PWD}:${PYTHONPATH}" \
python -c "import tensorrt_llm; print(tensorrt_llm.__version__)" || \
echo "Warning: Import test failed - this is expected if CUDA/TensorRT are not available"
}
package() {
cd "${_pkgname}-${pkgver}"
# Install the package using modern installer
python -m installer --destdir="$pkgdir" dist/*.whl
# Install licenses
install -Dm644 LICENSE "${pkgdir}/usr/share/licenses/${pkgname}/LICENSE"
# Install attribution files (x86_64 only)
if [ -f "ATTRIBUTIONS-CPP-x86_64.md" ]; then
install -Dm644 "ATTRIBUTIONS-CPP-x86_64.md" "${pkgdir}/usr/share/licenses/${pkgname}/ATTRIBUTIONS-CPP-x86_64.md"
fi
if [ -f "ATTRIBUTIONS-Python.md" ]; then
install -Dm644 "ATTRIBUTIONS-Python.md" "${pkgdir}/usr/share/licenses/${pkgname}/ATTRIBUTIONS-Python.md"
fi
# Install documentation
install -Dm644 README.md "${pkgdir}/usr/share/doc/${pkgname}/README.md"
}
|