import os import platform import sys import pkg_resources from setuptools import find_packages, setup def read_version(fname="whisper/version.py"): exec(compile(open(fname, encoding="utf-8").read(), fname, "exec")) return locals()["__version__"] requirements = [] if sys.platform.startswith("linux") and platform.machine() == "x86_64": requirements.append("triton==2.0.0") setup( name="openai-whisper", py_modules=["whisper"], version=read_version(), description="Robust Speech Recognition via Large-Scale Weak Supervision", long_description=open("README.md", encoding="utf-8").read(), long_description_content_type="text/markdown", readme="README.md", python_requires=">=3.8", author="OpenAI", url="https://github.com/openai/whisper", license="MIT", packages=find_packages(exclude=["tests*"]), install_requires=requirements + [ str(r) for r in pkg_resources.parse_requirements( open(os.path.join(os.path.dirname(__file__), "requirements.txt")) ) ], entry_points={ "console_scripts": ["whisper=whisper.transcribe:cli"], }, include_package_data=True, extras_require={"dev": ["pytest", "scipy", "black", "flake8", "isort"]}, )