rulka/pyproject.toml at main · ArtyomZemlyak/rulka · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
[project]
name = "rulka"
version = "1.8.0"
description = "Trackmania AI with reinforcement learning - Rulka fork"
license = {text = "MIT"}
requires-python = ">=3.10,<3.12"
readme = "README.md"
authors = [
    {name = "artyo"},
]

dependencies = [
    # Core ML (PyTorch version/source pinned under [tool.uv.sources])
    "torch>=2.7.0",
    "torchvision>=0.22.0",
    "torchaudio>=2.7.0",
    "torchrl>=0.6.0",

    # Scientific computing
    "numpy>=1.26.4,<2.0",  # numpy 2.x breaks numba
    "scipy>=1.14.0",
    "numba>=0.60.0",

    # Visualization & monitoring
    "matplotlib>=3.9.0",
    "tensorboard>=2.18.0",
    "prettytable>=3.11.0",
    "art>=6.3",

    # Image processing
    "opencv-python>=4.10.0",

    # Utilities
    "joblib>=1.4.0",
    "psutil>=6.0.0",
    "tqdm>=4.66.0",
    "setuptools",
    "pydantic>=2.0",
    "pydantic-settings>=2.0",
    "pyyaml>=6.0",
    "python-dotenv>=1.0",
    # BC pretrain (scripts/pretrain_bc.py) and Level-0 pretrain Lightning path
    "lightning>=2.0.0",

    # TrackMania specific
    "readwritememory",
    "tminterface",
    "pygbx",

    # Windows specific
    "pywin32>=306 ; sys_platform == 'win32'",
    "triton-windows>=3.2.0 ; sys_platform == 'win32'",  # Triton backend for torch.compile on Windows

    # Linux specific
    "python-libxdo>=0.1.0 ; sys_platform == 'linux'",
]

[project.optional-dependencies]
dev = [
    "ruff>=0.8.0",
]
# For trackmania_rl.pretrain (scripts/pretrain_visual_backbone.py): lightly (SimCLR), timm (future backbones). Lightning is a core dependency.
pretrain = [
    "lightly>=1.5.0",
    "timm>=1.0.0",
]
# Build docs: pip install -e ".[doc]". For IQN architecture diagrams, install system Graphviz (dot on PATH).
doc = [
    "sphinx>=8.0.0",
    "sphinx_rtd_theme>=3.0.0",
    "sphinxcontrib-youtube>=1.4.0",
]
# For scripts/dataset/convert_to_hf_dataset.py and push_to_hf.py: Hugging Face Hub upload.
hf = [
    "datasets>=3.0.0",
    "huggingface_hub>=0.25.0",
]
# PPO with optional Hugging Face vision backbone (ViT etc.): pip install -e ".[policy]"
# timm: Hub ids like timm/vit_* use transformers TimmWrapper and require timm at runtime.
policy = [
    "transformers>=4.45.0",
    "accelerate>=1.0.0",
    "timm>=1.0.0",
]

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

[tool.hatch.build.targets.wheel]
packages = ["trackmania_rl", "config_files"]

[tool.ruff]
include = ["pyproject.toml", "trackmania_rl/**/*.py", "scripts/**/*.py", "config_files/**/*.py"]
line-length = 140

[tool.uv]
# Windows x64 only: PyTorch cu128 nightlies on Linux pull triton with exact git builds that uv's
# indexes do not provide, so the lock becomes unsatisfiable for linux-* even when resolving on Windows.
# On Linux/WSL: install torch yourself (pip / pytorch.org) or maintain a separate lock for Linux.
environments = [
    "sys_platform == 'win32' and platform_machine == 'AMD64'",
]
# PyTorch nightly with CUDA 12.8 (e.g. RTX 50xx / sm_120)
index-strategy = "unsafe-best-match"

[[tool.uv.index]]
name = "pytorch-cu128-nightly"
url = "https://download.pytorch.org/whl/nightly/cu128"
explicit = true

[tool.uv.sources]
torch = { index = "pytorch-cu128-nightly" }
torchvision = { index = "pytorch-cu128-nightly" }
torchaudio = { index = "pytorch-cu128-nightly" }