pyproject.toml

[project]
name = "pyine"
version = "1.0.0"
readme = "README.md"
description = "Python Interpretation and Execution (PyINE) framework"
keywords = ["benchmarking", "python", "interpretation", "execution", "llm", "nlp"]
authors = [
    { name = "Pierre-Luc St-Charles", email = "pl.stcharles@lawzero.org" },
    { name = "Alessandro Palmas", email = "a.palmas@lawzero.org" }
]
license = "Apache-2.0"
requires-python = ">=3.12,<3.13"
dependencies = [
    # last review/update of versions: 2025-08-23
    "accelerate>=1.10.0,<2.0.0",
    "backoff>=2.2.1,<3.0.0",
    "blobfile>=3.0.0,<4.0.0",
    "datasets>=4.0.0,<5.0.0",
    "deepdiff>=8.6.0,<9.0.0",
    "deepspeed>=0.18.3",
    "evaluate>=0.4.5,<1.0.0",
    "filelock>=3.20.0,<4.0.0",
    "gitpython>=3.1.45,<4.0.0",
    "hydra-colorlog>=1.2.0,<2.0.0",
    "hydra-core>=1.3.2,<1.4.0",
    "hydra-joblib-launcher>=1.2.0,<2.0.0",
    "hydra-wandb-sweeper @ git+https://github.com/plstcharles-saifh/hydra-wandb-sweeper.git@0.0.6",
    "hydra-zen>=0.16.0,<1.0.0",
    "ipykernel>=7.1.0",
    "ipython>=8.37.0",
    "langchain-huggingface>=0.3.1,<1.0.0",
    "langchain>=0.3.27,<0.4.0",
    "langchain_deepseek>=0.1.4,<0.2.0",
    "langgraph>=0.3,<1.0",
    "Levenshtein>=0.27.1,<0.28.0",
    "lightning>=2.5.3,<2.6.0",
    "lmdb>=1.7.3,<2.0.0",
    "lz4>=4.4.4,<5.0.0",
    "matplotlib>=3.10.5,<4.0.0",
    "msgspec>=0.19.0,<0.20.0",
    "numpy<2.3.0",
    "openai>=1.101.0,<2.0.0",
    "orjson>=3.11.2,<3.12.0",
    "pandas-stubs~=2.3.3",
    "pandas>=2.3.2,<3.0.0",
    "peft>=0.17.1,<1.0.0",
    "pydantic>=2.11.7,<3.0.0",
    "pyright>=1.1.408",
    "python-dotenv>=1.1.1,<2.0.0",
    "python-minifier>=2.11.3,<3.0.0",
    "radon>=6.0.1",
    "rich>=14.1.0,<15.0.0",
    "ruff>=0.13.2,<1.0.0",
    "scikit-learn>=1.8.0",
    "scikit-learn>=1.8.0,<2.0.0",
    "scipy-stubs~=1.17.1",
    "scipy>=1.17.1,<2.0.0",
    "tensorboard>=2.20.0",
    "tiktoken>=0.11.0,<1.0.0",
    "torch>=2.8.0,<3.0.0",
    "tqdm>=4.67.1,<5.0.0",
    "transformers>=4.55.4,<5.0.0",
    "trl>=0.27.1",
    "unidiff>=0.7.5,<0.8.0",
    "wandb>=0.25.1,<1.0.0",
]

[project.optional-dependencies]
dev = [
    "adjusttext>=1.3.0",
    "coverage[toml]>=7.10.0",
    "ipython>=8.0.0,<9.0.0",
    "ipywidgets>=8.1.0,<9.0.0",
    "nbstripout>=0.8.0,<1.0.0",
    "notebook>=7.3.0,<8.0.0",
    "pre-commit>=4.3.0,<5.0.0",
    "pyright[nodejs]>=1.1.408",
    "pytest-asyncio>=1.1.0",
    "pytest-cov>=6.2.0",
    "pytest-mock>=3.14",
    "pytest-timeout>=2.4.0",
    "pytest>=8.0.0",
    "seaborn>=0.13.2",
    "wordcloud>=1.9.4",
]
vllm = [
    "trl[vllm]>=0.27.1; sys_platform != 'darwin'",  # can't install vLLM on MacOS
]
flash_attn = [
    "flash-attn>=2.8.3; sys_platform == 'linux'",  # CUDA-only, Linux
]
liger = [
    "liger-kernel>=0.6.4; sys_platform == 'linux'",  # CUDA-only, Linux
]
gpu_monitoring = [
    "nvidia-ml-py>=13.590.0; sys_platform == 'linux'",  # NVML bindings for GPU stats
]

[build-system]
requires = ["setuptools>=42", "wheel"]
build-backend = "setuptools.build_meta"

[tool.setuptools]
packages = ["pyine"]

[tool.setuptools.package-data]
"pyine" = ["**/*.yaml"]

[tool.pytest.ini_options]
testpaths = ["tests"]
markers = [
    "slow: marks tests as slow to run due to data/compute (deselect with '-m \"not slow\"')",
    "integration: marks end-to-end or cross-service tests that rely on external resources/APIs (deselect with '-m \"not integration\"')",
    "dataset: marks tests that require large local datasets (e.g. TACO) to be present (deselect with '-m \"not dataset\"')",
    "openai: marks tests that exercise the live OpenAI API and require network access plus OPENAI_API_KEY (deselect with '-m \"not openai\"')",
    "distributed: marks tests that require distributed/multi-process execution (deselect with '-m \"not distributed\"')",
]
filterwarnings = [
    # until hydra-zen fixes their pydantic v1 warnings, we need this (currently using 0.15.0)
    'ignore:The `__fields__` attribute is deprecated, use the `model_fields` class property instead\.:pydantic.warnings.PydanticDeprecatedSince20:hydra_zen\.structured_configs\._implementations',
    'ignore:.*hydra-zen.*pydantic v1.*:UserWarning',
    'ignore:.*:pydantic.warnings.UnsupportedFieldAttributeWarning',
    # pytorch 2.9+ deprecation warnings triggered by internal/dependency code
    'ignore:.*allow_tf32.*will be deprecated after Pytorch 2\.9.*:UserWarning',
    'ignore:.*pin_memory.*is deprecated.*:DeprecationWarning',
    'ignore:.*is_pinned.*is deprecated.*:DeprecationWarning',
    # SWIG deprecation warnings from dependencies
    'ignore:builtin type SwigPyPacked has no __module__ attribute:DeprecationWarning',
    'ignore:builtin type SwigPyObject has no __module__ attribute:DeprecationWarning',
]

[tool.coverage.run]
branch = true
parallel = true
relative_files = true
source = ["pyine"]
data_file = "logs/coverage/.coverage"
omit = [
  "tests/*",
  "**/__init__.py",
]

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

[tool.coverage.html]
directory = "logs/coverage/html"

[tool.coverage.xml]
output = "logs/coverage/coverage.xml"

# Legacy tool configs kept for nbqa compatibility
[tool.isort]
profile = "black"
line_length = 120
src_paths = ["pyine", "tests"]
known_first_party = ["pyine", "tests"]

[tool.nbqa]
files = "^notebooks/"
[tool.nbqa.addopts]
ruff = ["--line-length=120"]
ruff-format = ["--line-length=120"]

[tool.ruff]
line-length = 120
indent-width = 4

[tool.ruff.lint]
select = [
    # Core linting
    "E",   # pycodestyle errors (spacing, indentation, etc.)
    "F",   # pyflakes (unused imports, variables, etc.)
    "W",   # pycodestyle warnings
    # Code quality
    "I",   # isort (import sorting and organization)
    "B",   # flake8-bugbear (common bugs and design problems)
    "C4",  # flake8-comprehensions (mccabe; better list/dict comprehensions)
    "UP",  # pyupgrade (modern Python syntax)
    "RUF008", # mutable-dataclass-default
    "RUF012", # mutable-class-default
    # Type annotations
    "ANN", # flake8-annotations (enforce type hints)
    "TC",  # flake8-type-checking (type-checking imports)
    # Security and best practices
    "S",   # flake8-bandit (security issues)
    "N",   # pep8-naming (naming conventions)
    "RET", # flake8-return (return statement issues)
    "SIM", # flake8-simplify (code simplification)
]
ignore = [
    # Type annotation exceptions
    "ANN401", # any-type (typing.Any allowed in framework)
    # Security exceptions for common patterns
    "S101",   # assert (use of assert is allowed in framework, common in tests)
    "S603",   # subprocess-without-shell-equal...
    "S311",   # suspicious-non-cryptographic-....
    # Code unreadability
    "SIM102", # collapsible-if (allow nested `if`)
    "SIM108", # if-else-block-instead-of-... (allow ternary operator)
]
[tool.ruff.lint.isort]
known-third-party = ["wandb"]

[tool.ruff.format]
# Formatting style - matching Black formatting
indent-style = "space"
quote-style = "preserve"  # Don't enforce single vs double quotes
docstring-code-format = true  # Format code blocks in docstrings
skip-magic-trailing-comma = false  # Keep Black's trailing comma behavior
line-ending = "auto"  # Use system default line endings

[tool.uv.extra-build-dependencies]
flash-attn = ["torch"]

[tool.pyright]
typeCheckingMode = "strict"
include = ["pyine"]
exclude = ["tests"]
venvPath = "."
venv = ".venv"
# Overrides from strict defaults for data science workflows
reportMissingTypeStubs = false        # Ignore missing third-party type stubs
reportOptionalMemberAccess = false    # Allow optional member access (e.g., df.column)
reportOptionalSubscript = false       # Allow optional subscript access without None checks