Skip to article frontmatterSkip to article content
Site not loading correctly?

This may be due to an incorrect BASE_URL configuration. See the MyST Documentation for reference.

Dataset catalog

Use this page to choose a detector, input representation, dataset name, and compatible model recipe. The download guide covers published data, while the generation guide covers custom data production.

Choose an input representation

Track-and-cluster datasets (*_pf) are the supported default and contain substantially fewer input elements per event. Choose a hit dataset (*_hits) for studies that need tracker and calorimeter hits. Hit datasets use dedicated model recipes and require more memory; see Hit-based datasets.

Current recipes

The table describes the recipes on the current development branch. Dataset versions identify the data schema and production. The Python package and model checkpoints have their own versions, so record all three in derived work.

DetectorInput representationDataset namesRecipe versionCompatible model recipe
CMS Run 3, pileupTracks and calorimeter clusterscms_pf_ttbar, cms_pf_qcd, cms_pf_ztt3.2.0pyg-cms-v1
CMS Run 3, no pileupTracks and calorimeter clusterscms_pf_ttbar_nopu, cms_pf_qcd_nopu, cms_pf_ztt_nopu3.2.0pyg-cms-v1
CLD, 365 GeVTracks and calorimeter clusterscld_edm_ttbar_pf, cld_edm_ww_fullhad_pf, cld_edm_qq_pf, cld_edm_zz_pf3.2.1pyg-cld-v1 (defaults: ttbar, WW, qq)
CLD, 365 GeVTracker and calorimeter hitscld_edm_ttbar_hits, cld_edm_ww_fullhad_hits, cld_edm_qq_hits, cld_edm_zz_hits3.2.1pyg-cld-hits-v1 (defaults: ttbar, WW, qq)
CLICdet, 380 GeVTracks and calorimeter clustersclic_edm_ttbar_pf, clic_edm_ww_fullhad_pf, clic_edm_qq_pf3.2.1pyg-clic-v1
CLICdet, 380 GeVTracker and calorimeter hitsclic_edm_ttbar_hits, clic_edm_ww_fullhad_hits, clic_edm_qq_hits3.2.1pyg-clic-hits-v1

IDEA 0.1.0 datasets support pipeline validation with truth-seeded proxy tracks and an oracle reference. The table focuses on the supported training and research workflows. The IDEA dataset’s scope is recorded in Current capabilities.

Configuration partitions and event splits

Each recipe defines configuration partitions 1 through 10. A configuration partition contains its own train and test event splits: the number identifies the partition, while train or test identifies the event split. The default model recipes combine all ten partitions. Configuration 1 provides a one-partition subset for a small experiment; the configured training corpus spans all ten.

The on-disk layout is:

<data root>/<dataset name>/<configuration>/<version>/

For example:

data/tfds/tensorflow_datasets/cld/cld_edm_ttbar_pf/1/3.2.1/

Pass the detector-level directory containing the dataset-name directories as --data-dir; for this example it is data/tfds/tensorflow_datasets/cld.

Published availability

As checked on 14 September 2026, the public Hugging Face dataset repository contains configuration 1 of the CLD and CLIC datasets above at their current recipe versions. It also contains older versions and the experimental IDEA datasets. CMS 3.2.0 is currently available through the configured local production workflow.

Treat the live Hub tree as the source of truth for publication availability. The tfds_mapping, tfds_hit_mapping, and model dataset entries in particleflow_spec.yaml define what the current code expects. A usable dataset has a detector, name, configuration, and version that match in both contexts.

For detector-software and sample details, see CMS data or CLD and CLIC data. For scientific provenance and citations, use the publication map and the dataset record associated with the version you consume.