Add pyproject.toml with build-system metadata

flash_mla/__init__.py

@@ -1,4 +1,8 @@
-__version__ = "1.0.0"
+try:
+    from importlib.metadata import version as _get_version
+    __version__ = _get_version("flash_mla")
+except (ImportError, ModuleNotFoundError):
+    from flash_mla._version import __version__
 
 from flash_mla.flash_mla_interface import (
     get_mla_metadata,

flash_mla/_version.py

@@ -0,0 +1 @@
+__version__ = "1.0.0"

pyproject.toml

@@ -0,0 +1,34 @@
+[build-system]
+requires = ["setuptools>=61.0"]
+# torch is NOT listed above because pip's isolated build would pull a CPU-only
+# wheel from PyPI, which lacks CUDA support. FlashMLA requires a CUDA-capable
+# torch pre-installed in the target environment before building.
+build-backend = "setuptools.build_meta"
+
+[project]
+name = "flash_mla"
+dynamic = ["version"]
+description = "DeepSeek's library of optimized attention kernels, powering DeepSeek-V3 and DeepSeek-V3.2-Exp models"
+readme = "README.md"
+license = {text = "MIT"}
+authors = [
+    {name = "DeepSeek"},
+]
+requires-python = ">=3.8"
+dependencies = [
+    "torch",
+]
+classifiers = [
+    "Programming Language :: Python :: 3",
+    "License :: OSI Approved :: MIT License",
+    "Operating System :: OS Independent",
+]
+
+[project.urls]
+Homepage = "https://github.com/deepseek-ai/FlashMLA"
+Repository = "https://github.com/deepseek-ai/FlashMLA"
+
+# [tool.setuptools.dynamic] is intentionally omitted.
+# Version is computed in setup.py (base from flash_mla/_version.py + git hash).
+# The attr directive is unusable here: importing any flash_mla submodule
+# triggers __init__.py, which imports the C extension that is not yet built.

setup.py

@@ -53,6 +53,14 @@ def get_nvcc_thread_args():
 
 this_dir = os.path.dirname(os.path.abspath(__file__))
 
+# Read base version from _version.py without importing the flash_mla package.
+# Python always executes __init__.py first when importing any submodule,
+# and flash_mla/__init__.py imports the C extension which is not yet built.
+_ver_ns = {}
+with open(os.path.join(this_dir, "flash_mla", "_version.py")) as f:
+    exec(f.read(), _ver_ns)
+__base_version__ = _ver_ns.get("__version__", "1.0.0")
+
 if IS_WINDOWS:
     cxx_args = ["/O2", "/std:c++20", "/DNDEBUG", "/W0"]
 else:
@@ -144,7 +152,7 @@ def get_nvcc_thread_args():
 
 setup(
     name="flash_mla",
-    version="1.0.0" + rev,
+    version=__base_version__ + rev,
     packages=find_packages(include=['flash_mla']),
     ext_modules=ext_modules,
     cmdclass={"build_ext": BuildExtension},