pyproject.toml

[build-system]
requires = [
  "setuptools>=82",
  "setuptools_scm>=8"
]
build-backend = "setuptools.build_meta"

[project]
name = "spyre-inference"
description = "Next iteration of sendnn-inference on the torch-spyre stack"
readme = "README.md"
license = {text = "Apache-2.0"}
dependencies = [
    "torch-spyre",
    "vllm",

    # Listing as direct deps to ensure they're picked up from the correct index.
    # Versions intentionally left out because these are dependencies of dependencies.
    "torch",
    "torchvision",
]
requires-python = ">=3.11"
dynamic = ["version"]

[project.urls]
Homepage = "https://github.com/torch-spyre/spyre-inference"
Documentation = "https://torch-spyre.github.io/spyre-inference/"
Repository = "https://github.com/torch-spyre/spyre-inference"
Issues = "https://github.com/torch-spyre/spyre-inference/issues"

[project.entry-points."vllm.platform_plugins"]
spyre_inference = "spyre_inference:register"

[project.entry-points."vllm.general_plugins"]
spyre_inference_ops = "spyre_inference:register_ops"

[tool.setuptools.packages.find]
where = ["."]  # list of folders that contain the packages (["."] by default)
include = ["spyre_inference*"]  # package names should match these glob patterns (["*"] by default)
exclude = ["spyre_inference.testing*"]  # exclude testing module (now a separate package)
namespaces = false  # to disable scanning PEP 420 namespaces (true by default)

[tool.setuptools_scm]
# Version files are optional, this just marginally speeds up version inspection
# at runtime
version_file = "spyre_inference/_version.py"
# Version strings are formatted as `{version}+{local}`, e.g. "0.1.dev1+gb1a7032"
# Local versioning is incompatible with pypi, so we disable it by default.
local_scheme = "no-local-version"


[tool.uv]
default-groups = ["dev"]
override-dependencies = [
    # No triton support yet, don't allow any dependencies to pull it in
    "triton; sys_platform == 'never'",
    "arctic-inference; sys_platform == 'never'",
    "perf-analyzer; sys_platform == 'never'",

    # Skip packages on s390x and ppc64le that are expected to be pre-installed
    "vllm ; platform_machine not in 's390x, ppc64le'",
    "ray; platform_machine not in 's390x, ppc64le'",
    # uv doesn't resolve numba->llvmlite constraint correctly...
    #   if numba is updated, this version will likely need to be too
    "llvmlite==0.44.0; platform_machine not in 's390x, ppc64le'",
    "pyarrow; platform_machine not in 's390x, ppc64le'",
    # This torch version must match the one below in `build-constraint-dependencies`
    "torch==2.10.0 ; platform_machine not in 's390x, ppc64le'",

    # torch-spyre declares a dependency on numpy>2.3, conflicting with numba needed by vLLM
    # The range here should match the numba dependency
    "numpy>=1.24,<2.3",

    # opencv-python-headless==4.13.0.90 has a packaging bug that incorrectly depends on libxcb.so.1,
    # causing import failure on headless Linux systems.
    "opencv-python-headless==4.12.0.88",

    # setuptools is specified as a runtime dep by vllm, make sure it's patched for CVEs
    "setuptools>=82",

    # requests is used by many dependencies, make sure it's patched for CVEs
    "requests>=2.32.4",

]
# This adds constraints to all dependent build environments, which will ensure everything is built
# with the same version of torch. This CANNOT conflict with a package's existing build dependencies
build-constraint-dependencies = ["torch==2.10.0"]
# This ensures that we build the cpu backend for vLLM
extra-build-variables = { vllm = { VLLM_TARGET_DEVICE = "cpu" } }


# vLLM and torch-spyre must be pulled from github to be built from source
# NB: torch-spyre can only be compiled where `sendnn` is available
[tool.uv.sources]
vllm = { git = "https://github.com/vllm-project/vllm", rev = "v0.19.1" }
torch-spyre = { git = "https://github.com/torch-spyre/torch-spyre", rev = "bdd54af5549532e4ff7a21c9787b1397aeb85c75" }
torch = [
  { index = "pytorch-cpu" },
]
torchvision = [
  { index = "pytorch-cpu" },
]
# Testing plugin package
spyre-testing-plugin = { workspace = true }

[tool.uv.workspace]
members = ["tests/plugin"]

# vllm specifies "common" build dependencies in pyproject.toml, while specific requirements for
# cpu builds are specified here:
# See https://github.com/vllm-project/vllm/blob/v0.15.1/requirements/cpu-build.txt
# In the future, we may need to use this section to specify extra dependencies required to build
# vllm from source for the cpu backend.
# NB: This section CANNOT be used to override dependencies to a conflicting version. For example,
# you cannot add torch==2.9.1 if vllm's build dependencies already specify torch>=2.10.0
# [tool.uv.extra-build-dependencies]
# vllm = [
#     "ninja",
# ]

# The pytorch-cpu index here ensures we always pull the cpu version of torch everywhere
# The `explicit=true` setting means this index will only be used for packages that we explicitly set
# it for in `tool.uv.sources`. This is important to avoid requiring the use of 
# `--index-strategy unsafe-best-match`
[[tool.uv.index]]
name = "pytorch-cpu"
url = "https://download.pytorch.org/whl/cpu"
explicit = true

[tool.ty.rules]
possibly-missing-attribute = "ignore"
# TODO: re-enable once vllm/torch-spyre imports resolve in CI.
unresolved-import = "ignore"
# TODO: re-enable after fixing subclass attribute lookups in custom_ops/linear.py.
unresolved-attribute = "ignore"

[tool.ty.src]
exclude = ["spyre_inference/__init__.py"]

[tool.ruff]
# Use widescreen monitors 😎😎😎
line-length = 100

[tool.ruff.lint.per-file-ignores]
"spyre_inference/version.py" = ["F401"]
"spyre_inference/_version.py" = ["ALL"]
"spyre_inference/custom_ops/__init__.py" = ["F401"]

[tool.ruff.lint]
select = [
    # pycodestyle
    "E",
    # Pyflakes
    "F",
    # pyupgrade
    "UP",
    # flake8-bugbear
    "B",
    # flake8-simplify
    "SIM",
    # isort
    # "I",
    "G",
]
ignore = [
    # star imports
    "F405", "F403",
    # lambda expression assignment
    "E731",
    # Loop control variable not used within loop body
    "B007",
    # f-string format
    "UP032",
    # % formatters
    "UP031",
    # ternary operators
    "SIM108",


    # TODO: Fix all of these in code instead

    # Type unions
    "UP007",
    # Callable
    "UP035",
    # Zip
    "B905",
]

[tool.codespell]
ignore-words-list = "dout, te, indicies, subtile, ElementE"
skip = "*.svg,./tests/hf_cache.json"
#skip = "./tests/models/fixtures,./tests/prompts,./benchmarks/sonnet.txt,./tests/lora/data,./build"

[tool.isort]
use_parentheses = true
skip_gitignore = true

[tool.pytest.ini_options]
pythonpath = ["."]
asyncio_default_fixture_loop_scope = "function"
filterwarnings = [
    "ignore::_pytest.warning_types.PytestUnknownMarkWarning"
]

# --8<-- [start:test-markers-definition]
markers = [
    "spyre: Tests that require a Spyre device",
    "upstream: Tests coming from upstream vLLM",
]
# --8<-- [end:test-markers-definition]

[tool.markdownlint]
#MD013.line_length = 100
MD013 = false # line-length
MD033 = false # inline-html
MD038 = false # no-space-in-code
MD041 = false # first-line-h1
MD046 = false # code-block-style
MD024.allow_different_nesting = true # no-duplicate-headers
MD007.indent = 4 # ul-indent
MD037 = false # no-space-in-emphasis (allow MkDocs Admonitions)

[tool.pymarkdown]
plugins.md013.enabled = false # line-length
plugins.md033.enabled = false # inline-html
plugins.md041.enabled = false # first-line-h1
plugins.md046.enabled = false # code-block-style
plugins.md024.allow_different_nesting = true # no-duplicate-headers
plugins.md007.enabled = true
plugins.md007.indent = 4

[dependency-groups]
dev = [
    "pytest",
    "pyyaml",
    "spyre-testing-plugin",
]