Configuration Templates¶

There are two kinds of ready-made config, and which one you want depends on whether you are bringing your own model or running one of ours.

	Model examples	Blank templates
Run against	a pretrained model from the zoo	your own checkpoint and data
Paths	already resolved — nothing to edit	you fill them in
Where	installed with the package	`docs/cfg_examples/` in the repository
Start with	`MolCraftDiff generate examples/<name>.yaml`	copy, edit, run

New here? Use a model example — it runs as-is. Reach for a template once you have trained something of your own.

Model examples (ready to run)¶

Every model in the zoo ships a config that already points at its weights and data, so it runs unedited on any machine:

MolCraftDiff zoo fetch --model kgdiff
MolCraftDiff generate examples/kgdiff_generate.yaml

examples/ is inside the installed package, not a folder in your working directory — so this works from anywhere. List what is there, and copy one out if you want to keep changes:

MolCraftDiff zoo config                              # list them
MolCraftDiff zoo config kgdiff_generate.yaml .       # copy into the current dir

A local file of the same name takes precedence over the bundled one, so an edited copy is picked up automatically.

For one-off changes, skip copying and override on the command line:

MolCraftDiff generate examples/kgdiff_generate.yaml \
    interference.num_generate=100 interference.output_path=my_run

See the Model Zoo for the full workflow, and Model Architectures for choosing a model.

Blank templates (bring your own model)¶

These live in the repository under docs/cfg_examples/. Copy the relevant file, fill in your checkpoint and data paths, and run it. The packaged Hydra defaults they build on are under src/MolecularDiffusion/configs/.

Training¶

Template	Purpose
`docs/cfg_examples/train_diffusion.yaml`	Train an EDM diffusion model from scratch.
`docs/cfg_examples/finetune_diffusion.yaml`	Fine-tune from a pretrained diffusion checkpoint.
`docs/cfg_examples/train_regressor.yaml`	Train a property regressor.
`docs/cfg_examples/train_guidance.yaml`	Train a guidance (property) model.

MolCraftDiff train docs/cfg_examples/train_diffusion

Generation¶

Template	Mode (`interference.task_type`)
`docs/cfg_examples/gen_unconditional.yaml`	`unconditional`
`docs/cfg_examples/gen_cfg.yaml`	`cfg` — classifier-free guidance
`docs/cfg_examples/gen_gradient_guidance.yaml`	`gg` — gradient guidance
`docs/cfg_examples/gen_hybrid_cfg_gg.yaml`	`cfggg` — hybrid CFG + GG
`docs/cfg_examples/gen_inpaint.yaml`	`inpaint` — structure inpainting
`docs/cfg_examples/gen_outpaint.yaml`	`outpaint` — fragment extension

MolCraftDiff generate docs/cfg_examples/gen_cfg

Note

The structure-guided templates (gen_inpaint, gen_outpaint) need a reference .xyz. Three are shipped with the zoo — fetch them with MolCraftDiff zoo fetch inputs/templates — or point condition_configs.reference_structure_path at your own.

Config-driven commands accept Hydra-style dotted overrides, e.g. MolCraftDiff generate docs/cfg_examples/gen_cfg interference.num_generate=200. See Tutorials for end-to-end walkthroughs.