A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 17, 06:45 UTC
Build CUDA11.8 wheels for release (#1596)
Woosuk Kwon committed
Nov 9, 2023 at 03:52 UTC
fd58b73a40d937ea6d2c55e5a8147cc0a605efe2
2 files changed
+14
-3
.github/workflows/publish.yml
+1
-1
@@ -50,7 +50,7 @@ jobs:
50
os: ['ubuntu-20.04']
51
python-version: ['3.8', '3.9', '3.10', '3.11']
52
pytorch-version: ['2.1.0']
53
- cuda-version: ['12.1']
53
+ cuda-version: ['11.8', '12.1']
54
55
steps:
56
- name: Checkout
setup.py
+13
-2
@@ -12,6 +12,8 @@ from torch.utils.cpp_extension import BuildExtension, CUDAExtension, CUDA_HOME
12
13
ROOT_DIR = os.path.dirname(__file__)
14
15
+MAIN_CUDA_VERSION = "12.1"
16
+
17
# Supported NVIDIA GPU architectures.
18
SUPPORTED_ARCHS = {"7.0", "7.5", "8.0", "8.6", "8.9", "9.0"}
19
@@ -225,7 +227,7 @@ def get_path(*filepath) -> str:
227
return os.path.join(ROOT_DIR, *filepath)
228
229
228
-def find_version(filepath: str):
230
+def find_version(filepath: str) -> str:
231
"""Extract version information from the given filepath.
232
233
Adapted from https://github.com/ray-project/ray/blob/0b190ee1160eeca9796bc091e07eaebf4c85b511/python/setup.py
@@ -238,6 +240,15 @@ def find_version(filepath: str):
240
raise RuntimeError("Unable to find version string.")
241
242
243
+def get_vllm_version() -> str:
244
+ version = find_version(get_path("vllm", "__init__.py"))
245
+ cuda_version = str(nvcc_cuda_version)
246
+ if cuda_version != MAIN_CUDA_VERSION:
247
+ cuda_version_str = cuda_version.replace(".", "")[:3]
248
+ version += f"+cu{cuda_version_str}"
249
+ return version
250
+
251
+
252
def read_readme() -> str:
253
"""Read the README file if present."""
254
p = get_path("README.md")
@@ -256,7 +267,7 @@ def get_requirements() -> List[str]:
267
268
setuptools.setup(
269
name="vllm",
259
- version=find_version(get_path("vllm", "__init__.py")),
270
+ version=get_vllm_version(),
271
author="vLLM Team",
272
license="Apache 2.0",
273
description=("A high-throughput and memory-efficient inference and "