File size: 3,738 Bytes
9a6ea76 | 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 | [build-system]
requires = ["setuptools>=68", "wheel"]
build-backend = "setuptools.build_meta"
[project]
name = "jlens-materials"
version = "0.1.0"
description = "Jacobian-lens interpretability toolkit for materials science and protein prompts — fit/apply the lens, plot publication figures, run causal swaps, and write an LLM-authored LaTeX report."
readme = "README.md"
requires-python = ">=3.10"
license = { text = "Apache-2.0" }
authors = [{ name = "Markus J. Buehler", email = "mbuehler@mit.edu" }]
keywords = [
"interpretability", "jacobian-lens", "mechanistic-interpretability",
"transformers", "materials-science", "protein", "fracture-mechanics",
]
classifiers = [
"Programming Language :: Python :: 3",
"License :: OSI Approved :: Apache Software License",
"Intended Audience :: Science/Research",
"Topic :: Scientific/Engineering :: Artificial Intelligence",
]
# Core runtime — everything the fit -> apply -> figures -> analysis -> report
# pipeline needs. Providers (anthropic/openai) are included so the LLM analysis
# works out of the box; the --offline path does not require them at runtime.
dependencies = [
"torch>=2.2",
"transformers>=4.45", # Gemma-3/4 need a recent release (dev env: 5.9)
"accelerate>=0.30", # device_map + efficient model loading
"sentencepiece>=0.2", # Gemma / Llama / Qwen tokenizers
"datasets>=2.19", # wikitext prompts for lens fitting (--fit)
"huggingface_hub>=0.30", # private Hub storage for fitted lens bundles
"numpy>=1.24",
"matplotlib>=3.7",
"wordfreq==3.1.1", # frozen English-frequency screen for 31B SI streams
"pillow>=10.0", # GIF export for the "thinking" animation
"tqdm>=4.66", # progress bars for the slow lens fit
"anthropic>=0.40", # analysis provider: Claude (Messages API)
"openai>=1.66", # analysis provider: OpenAI (Responses API)
]
[project.optional-dependencies]
# Everything above already covers a full run; `dev` adds tooling only.
dev = ["ruff>=0.6", "ipython>=8.0", "pytest>=8.0"]
[project.urls]
Homepage = "https://github.com/lamm-mit/Substrates"
Repository = "https://github.com/lamm-mit/Substrates"
"Jacobian lens (upstream)" = "https://github.com/lamm-mit/jacobian-lens"
# ---------------------------------------------------------------------------
# Layout: the toolkit modules live flat in jlens_materials/ and import each
# other by bare name (import matviz, ...). We expose them as top-level modules
# and ship the vendored `jlens` (with its data/ resources) as a package, so an
# editable install makes `import run_lens`, `import matviz`, `import jlens`
# resolve from anywhere — while the scripts still run in-place from the dir.
# ---------------------------------------------------------------------------
[tool.setuptools]
py-modules = [
"analyze", "animate", "artifacts", "auto_swap", "compare", "domain_prompts",
"lens_hub", "llm", "matviz", "paper_protocol", "report", "report_latex", "run_lens", "swap",
]
packages = ["jlens", "explorer"]
[tool.setuptools.package-dir]
"" = "jlens_materials"
"jlens" = "jlens_materials/_vendor_jlens/jlens"
[tool.setuptools.package-data]
jlens = ["data/*"]
explorer = ["README.md", "launch.sh", "static/*", "skills/**/*"]
# Convenience entry points — same as `python <module>.py`, runnable anywhere.
[project.scripts]
jlens-run = "run_lens:main"
jlens-analyze = "analyze:main"
jlens-report = "report_latex:main"
jlens-report-md = "report:main"
jlens-compare = "compare:main"
jlens-animate = "animate:main"
jlens-swap = "swap:main"
jlens-auto-swap = "auto_swap:main"
jlens-artifacts = "artifacts:main"
jlens-explorer = "explorer.server:main"
|