Molab dependencies

[liquidcarbon]

In molab, there's a pretty long list of base dependencies, but if you download the notebook, they do not show up.

Should deps go into PEP 723 inline metadata?
Should there be a button to copy/export requirements or pyproject file?

Basically, how can it be made easy to recreate the notebook on a different machine? It's the same question I've always had for Colab.

root@modal:/marimo# cat pyproject.toml

[project]
dependencies = [
  "marimo[recommended,lsp]==0.14.12",
  "jax==0.6.2",
  "torch==2.7.1",
  "matplotlib>=3.8",
  "plotly>=5.21",
  "numpy>=2.0",
  "umap-learn>=0.5.8",
  "pymde>=0.2.3",
  "scikit-learn>=1.6.0",
  "spacy>=3.7.5",
  "nltk>=3.9.1",
  "transformers>=4.48.0",
  "sentence-transformers>=4.0.0",
  "xarray>=2025.1.0",
  "cvxpy>=1.5.0",
  "pandas>=2.0",
  "polars>=1.25",
  "huggingface-hub>=0.32.0",
  "geopandas>=1.0.0",
  "pyspark>=3.4.0",
  "tensorflow>=2.18.0",
  "xgboost>=2.1.4",
  "datasets>=2.14.4",
  "openai>=1.75.0",
  "wandb>=0.19",
  "pillow>=10.3",
  "lancedb>=0.20.0",
  "chromadb>=1.0",
]
name = "marimo-kernel"
# tensorflow not compatible with 3.13
requires-python = ">=3.12"
version = "0.1.0"

[tool.marimo.package_management]
manager = "uv"

[tool.uv]
index-url = "https://pypi.org/simple"

[tool.marimo.runtime]
auto_instantiate = false

[tool.marimo.ai.open_ai]
api_key = "unused"
model = "molab"
base_url = "https://ai.marimo.app/molab/v1"

[dmadisetti]

If you install deps, they will now persist in molab. You can manually uv sync, but as of right now- there isn't a cohesive sandboxed environment to used.

Good discussion though, I think this is still worth thinking about- but there are some technical/ requirements hurdles that make this not totally trivial to support