AssayLoop/pyproject.toml at main · Genentech/AssayLoop · 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
[project]
name = "assayloop"
version = "0.1.0"
description = "Sequential experimental design for genome-scale CRISPR screens: LLM warm-start handing off to a trained ranker."
readme = "README.md"
authors = [
    {name = "Edward De Brouwer", email = "debroue1@gene.com"},
    {name = "Carl Edwards", email = "edwardc24@gene.com"},
]
urls = {Homepage = "https://genentech.github.io/AssayLoop", Repository = "https://github.com/Genentech/AssayLoop"}
requires-python = ">=3.11"
license = {file = "LICENSE"}
dependencies = [
    # Screen corpus (Genentech/assaybench on the Hub) and ranking metrics.
    "assaybench>=0.2.0",
    # Core scientific stack. scipy and tqdm are imported directly (12 and 2
    # modules); they used to arrive transitively via scikit-learn/transformers,
    # which is how assaybench 0.1.0 shipped an ImportError.
    "numpy>=1.24",
    "pandas>=2.0",
    "scipy>=1.10",
    "scikit-learn>=1.4",
    "tqdm>=4.66",
    "pyyaml>=6.0",
    "python-dotenv>=1.0",
    # Hugging Face datasets: the public screen corpus loader.
    "datasets>=4.0.0",
    # LLM clients. `openai` backs the text-embedding-3-small DESC token
    # (amortized/text_embed.py) as well as the vLLM path.
    "openai>=1.40",
    # >=0.107: the agent_ranker baseline uses `anthropic.beta_tool` and
    # `client.beta.messages.tool_runner`, which 0.40 predates. This is the
    # version the agent runs were verified against.
    "anthropic>=0.107",
    # Baselines collector (DSPy-based, optional path)
    "dspy>=2.5",
    "requests>=2.28",
    # Plotting: every figure script under src/assayloop/scripts/.
    "matplotlib>=3.8",
    # CLI niceties
    "typer>=0.12",
    "rich>=13.0",
    "wandb>=0.27.1",
    "umap-learn>=0.5.12",
    "transformers>=5.10.2",
]

[project.optional-dependencies]
# torch is a 500 MB+ install and every module that needs it imports it
# lazily, so it is opt-in:
#   uv sync --extra torch
# Required for AssayFormer (train-ranker / eval-ranker / the amortized
# package) and for the Bayesian MLP baseline (`--model bayesian_mlp`).
torch = ["torch>=2.3"]
# Deprecated alias for the same extra.
mlp = ["torch>=2.3"]
# Small web service used by the public Try It page. Torch remains a separate
# extra so ordinary installs do not acquire the 500 MB+ runtime.
demo = [
    "fastapi>=0.110",
    "huggingface-hub>=0.30",
    "uvicorn[standard]>=0.29",
]
# Export/verification dependencies for deployment/export_onnx.py. These are
# build-time tools only; browser inference uses ONNX Runtime Web.
onnx = [
    "onnx>=1.17,<2",
    "onnxruntime>=1.20,<2",
]

[project.scripts]
assayloop = "assayloop.cli:app"

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

[tool.hatch.build.targets.wheel]
packages = ["src/assayloop"]

[tool.uv]
# uv's cache and this project live on different filesystems here, so hardlinks
# aren't possible; copy instead of emitting a warning on every sync.
link-mode = "copy"

[tool.uv.sources]
# PyPI's default torch wheel is built for CUDA 13.0 (+cu130), which the GPU
# hosts' CUDA 12.8 drivers cannot run. Pin torch to the cu128 build index
# (newest is 2.11.0+cu128); cu128 runtimes also run on newer (13.x) drivers.
torch = { index = "pytorch-cu128" }

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

[dependency-groups]
dev = [
    "pytest>=8.0",
    "ruff>=0.5",
    "jupyterlab>=4.0",
    "ipykernel>=6.29",
]