[project]
name = "sparkle-engine"
version = "0.1.0"
description = "Sparkle：Apple Silicon 上流式 MoE 推理（Qwen3-Next-80B 8bit）"
readme = "README.md"
# 仓库自带 native_moe_ext*.cpython-314*.so，须与扩展 ABI 一致
requires-python = "==3.14.*"
dependencies = [
    "fastapi>=0.115",
    "mlx>=0.31",
    "mlx-lm>=0.31",
    "numpy>=2.0",
    "textual>=0.80",
    "uvicorn>=0.30",
]

[project.scripts]
sparkle = "mlx_streaming.cli:main"

# 仅重编 native / 跑单测时需要；日常推理用已有 .venv + .so 即可
[dependency-groups]
dev = [
    "httpx>=0.27",
    "nanobind>=2.12.0",
    "pytest>=8",
    "pytest-asyncio>=0.23",
]

[[tool.uv.index]]
# 清华 PyPI 镜像
url = "https://pypi.tuna.tsinghua.edu.cn/simple"
default = true

[build-system]
requires = ["hatchling"]
build-backend = "hatchling.build"

[tool.hatch.build.targets.wheel]
packages = ["mlx_streaming"]

[tool.pytest.ini_options]
addopts = "-q"
testpaths = ["mlx_streaming/tests"]
asyncio_mode = "auto"