pyproject.toml

[build-system]
requires = [
  "setuptools>=48",
  "setuptools_scm[toml]>=6.3.1",
  # protobuf-generation tools
  "grpcio-tools==1.62.1",
  "mypy-protobuf==3.5.0",
  "types-protobuf>=3.20.4"
]
build-backend = "setuptools.build_meta"

[project]
name = "vllm-tgis-adapter"
description = "vLLM adapter for a TGIS-compatible grpc server"
readme = "README.md"
license = {text = "Apache 2"}
authors = [{name = "Daniele Trifirò", email = "dtrifiro@redhat.com"}]
classifiers = [
  "Programming Language :: Python :: 3",
  "Programming Language :: Python :: 3.9",
  "Programming Language :: Python :: 3.10",
  "Programming Language :: Python :: 3.11",
  "Programming Language :: Python :: 3.12",
  "Development Status :: 4 - Beta"
]
requires-python = ">=3.9"
dynamic = ["version"]
dependencies = [
  "vllm>=0.6.2",
  "prometheus_client==0.21.0",
  "grpcio==1.67.0",
  "grpcio-health-checking==1.62.2",
  "grpcio-reflection==1.62.2",
  "accelerate==0.34.2",
  "hf-transfer==0.1.8",
  # additional dependencies for OpenTelemetry tracing
  "opentelemetry-sdk>=1.26.0,<1.28.0",
  "opentelemetry-api>=1.26.0,<1.28.0",
  "opentelemetry-exporter-otlp>=1.26.0,<1.28.0",
  "opentelemetry-semantic-conventions-ai>=0.4.1,<0.5.0"
]

[project.urls]
Issues = "https://github.com/opendatahub-io/vllm-tgis-adapter/issues"
Source = "https://github.com/opendatahub-io/vllm-tgis-adapter"

[project.scripts]
grpc_healthcheck = "vllm_tgis_adapter.healthcheck:cli"
model-util = "vllm_tgis_adapter.tgis_utils.scripts:cli"
text-generation-server = "vllm_tgis_adapter.tgis_utils.scripts:cli"
convert_pt_to_prompt = "vllm_tgis_adapter.tgis_utils.convert_pt_to_prompt:cli"

[project.optional-dependencies]
tests = [
  "pytest==8.3.3",
  "pytest-cov==5.0.0",
  "pytest-mock==3.14.0",
  "pytest-asyncio==0.24.0",
  "requests==2.32.3"
]
dev = [
  "vllm_tgis_adapter[tests]",
  "ruff==0.6.9",
  "mypy==1.12.1",
  "types-protobuf==5.28.0.20240924",
  "types-requests==2.32.0.20240914"
]
flash_attn = [
  # it's easier to install flash-attn from wheel rather than like this as extra
  # "https://github.com/Dao-AILab/flash-attention/releases/download/v2.5.6/flash_attn-2.5.6+cu118torch2.0cxx11abiFALSE-cp311-cp311-linux_x86_64.whl",
  "flash-attn==2.6.3",
  "packaging",  # FIXME: temporary, until https://github.com/Dao-AILab/flash-attention/pull/937 is released
  "ninja"
]

[tool.setuptools_scm]
version_file = "src/vllm_tgis_adapter/_version.py"

[tool.setuptools.packages.find]
where = ["src"]
namespaces = false

[tool.setuptools.package-data]
vllm_tgis_adapter = [
  "**/*.proto"
]

[tool.pytest.ini_options]
addopts = "-ra -k \"not hf_data\""
markers = [
  "hf_data: marks tests that download data from HF hub (deselect with '-m \"not hf_data\"')"
]

[tool.coverage.run]
branch = true
source = ["vllm_tgis_adapter", "tests"]

[tool.coverage.paths]
source = ["src", "*/site-packages"]

[tool.coverage.report]
show_missing = true
exclude_lines = [
  "pragma: no cover",
  "if __name__ == .__main__.:",
  "if typing.TYPE_CHECKING:",
  "if TYPE_CHECKING:",
  "raise NotImplementedError",
  "raise AssertionError",
  "@overload"
]

[tool.mypy]
# Error output
show_column_numbers = true
show_error_codes = true
show_error_context = true
show_traceback = true
pretty = true
check_untyped_defs = false
# Warnings
warn_no_return = true
warn_redundant_casts = true
warn_unreachable = true
files = ["src", "tests"]

[[tool.mypy.overrides]]
ignore_missing_imports = true
module = [
  "grpc"
]

[tool.codespell]
ignore-words-list = " "

[tool.ruff]
output-format = "full"
show-fixes = true

[tool.ruff.lint]
ignore = [
  "S101",  # assert
  "PLR2004",  # magic-value-comparison
  "PLW2901",  # redefined-loop-name
  "ISC001",  # single-line-implicit-string-concatenation
  "SIM105",  # suppressible-exception
  "SIM108",  # if-else-block-instead-of-if-exp
  "D100",  # Missing docstring in public module
  "D101",  # Missing docstring in public class
  "D102",  # Missing docstring in public method
  "D103",  # Missing docstring in public function
  "D104",  # Missing docstring in public package
  "D107",  # Missing docstring in init
  "EM102",  # Exception must not use a string literal, assign to variable first
  "D203",  # one blank line before class
  "D213",  # multi-line-summary-second-line
  "PTH123",  # open() should be replaced by Path.open
  "ANN204",  # missing  issing return type annotation for special method `__init__`
  "EM101",  # Exception must not use a string literal, assign to variable first
  "ANN101",  # missing type annotation for `self` (deprecated, will be removed)
  "TRY003",  # Avoid specifying long messages outside the exception class
  "TD",  # todo messages
  "FIX",  # todo messages
  "ERA001",  # commented out code
  # formatting
  "COM812",
  "RET504"  # Unnecessary assignment to `args` before `return` statement
]
select = ["ALL"]

[tool.ruff.lint.per-file-ignores]
"**/__init__.py" = ["F401"]
"*src/vllm_tgis_adapter/grpc/grpc_server.py" = [
  "N802",  # name should be lowercase
  "ERA001"  # Found commented-out code
]
"src/vllm_tgis_adapter/healthcheck.py" = ["T201"]
"src/vllm_tgis_adapter/_version.py" = ["ALL"]
"tests/**" = ["S", "ARG001", "ARG002", "ANN", "PT019", "FBT001", "FBT002"]
"tests/utils.py" = ["T201"]
"setup.py" = [
  "T201",  # print() use
  "S603",  # subprocess call: check for execution of untrusted input
  "D",  # docs
  "ANN",  # annotations
  "TRY003"  # long messages outside exception class
]
"examples/*" = [
  "T201",
  "D",
  "ANN",
  "TRY003",
  'INP001',
  "PTH109",
  "EM101",
  "BLE001",
  "PLR0913"
]

[tool.ruff.lint.flake8-type-checking]
strict = true

[tool.ruff.lint.isort]
known-first-party = ["vllm_tgis_adapter"]