Data and Models

Data Contract

User-facing datasets are prompt-first. Committed prompt lists live under datasets/ (for example datasets/pickscore/train.txt), while larger datasets should be symlinked or mounted under datasets/ and passed through DATA_PATH or the recipe's data_source block.

bash examples/run_experiment_single_node.sh diffusion/sd3/sd3_trainside \
  data_source.args.run.data_path=/abs/path/train.json

See Data Preparation for the prompt-file formats and the per-prompt schema. Use the dataset-prep extra for repository converters; follow the nearest datasets/<name>/README.md for its output schema and canonical training recipe.

Runtime Model Code

Model implementation packages live under unirl/models/. Use the generated Model Package README for package layout and extension contracts.

Local Checkpoint Mounts

The repository-root models/ directory is for local artifacts (e.g. models/local/), not Python model code. Experiment YAMLs and model configs provide HuggingFace fallbacks through Hydra environment interpolation, for example:

pretrained_model_ckpt_path: ${oc.env:PRETRAINED_MODEL,stabilityai/stable-diffusion-3.5-medium}

Do not commit large model weights such as .bin, .safetensors, .pt, .pth, or .ckpt.