MolecularDiffusion.cli.zoo

Zoo CLI: discover, fetch and register pretrained weights and datasets.

MolCraftDiff zoo list shows what exists, fetch pulls only what you ask for, and add registers a newly integrated model so the zoo stays current instead of turning back into a pile of scattered files.

Attributes

logger

Functions

add_cmd(name, ckpt, data_paths, config_paths, variant, ...)	Register a newly integrated model's assets in zoo.yaml.
bundle_cmd(out, archive)	Build a self-contained, citable snapshot of the zoo.
config_cmd(name, dest)	Copy a bundled example config out so you can edit it.
fetch_cmd(names, model_name, config_path, fetch_all, ...)	Download assets into the cache.
info_cmd(target)	Show details for MODEL or MODEL/VARIANT.
list_cmd(tag, family, fetched, as_json, as_data)	List models in the zoo, or corpora with --data.
path_cmd(name)	Print an asset's local path, for $(...) interop.
recipe_cmd(name)	Show how to build a non-redistributable asset locally.
verify_cmd(names, verify_all)	Re-hash cached assets, and check the manifest is complete.
zoo()	Pretrained weights, datasets and example configs.

Module Contents

MolecularDiffusion.cli.zoo.add_cmd(name, ckpt, data_paths, config_paths, variant, task_type, family, tag)

Register a newly integrated model’s assets in zoo.yaml.

Hashes every file and writes the entry, leaving license and redistribute as TODO for you to fill in – zoo verify then fails until they are, which is what stops the zoo drifting.

MolecularDiffusion.cli.zoo.bundle_cmd(out, archive)

Build a self-contained, citable snapshot of the zoo.

Lays assets out in the CACHE layout, not the HuggingFace one, so the unpacked archive is usable with a single export MOLCRAFT_ASSETS=$(pwd) – no unpacking logic and no special format. Only redistribute: true assets are included; the manifest travels with them so the archive records exactly what it holds.

MolecularDiffusion.cli.zoo.config_cmd(name, dest)

Copy a bundled example config out so you can edit it.

With no NAME, lists what is available. The bundled configs live inside the installed package, so copying one by hand means finding that directory first – this does it for you. A config of the same name in your working directory takes precedence over the bundled one.

MolecularDiffusion.cli.zoo.fetch_cmd(names, model_name, config_path, fetch_all, force, dry_run)

Download assets into the cache.

Nothing is downloaded unless you ask for it. Pick a granularity: a single asset, one model, everything one config needs, or –all.

MolecularDiffusion.cli.zoo.info_cmd(target)

Show details for MODEL or MODEL/VARIANT.

MolecularDiffusion.cli.zoo.list_cmd(tag, family, fetched, as_json, as_data)

List models in the zoo, or corpora with --data.

MolecularDiffusion.cli.zoo.path_cmd(name)

Print an asset’s local path, for $(…) interop.

MolecularDiffusion.cli.zoo.recipe_cmd(name)

Show how to build a non-redistributable asset locally.

MolecularDiffusion.cli.zoo.verify_cmd(names, verify_all)

Re-hash cached assets, and check the manifest is complete.

Exits non-zero on any mismatch or any unaudited redistribute: TODO, so a half-registered model fails just check instead of being discovered a year later.

MolecularDiffusion.cli.zoo.zoo()

Pretrained weights, datasets and example configs.

MolecularDiffusion.cli.zoo.logger