Datasets & models

Where the data and the models live: open IE datasets first, then the general ML sources, then pretrained-model registries. Same rules as the rest of the tools section: each entry is dated and gets a one-line verdict.

Datasets

IE data:

• Industrial Ecology Freiburg data commons: open IE datasets (stocks, flows, lifetimes) collected by Stefan Pauliuk’s group; the closest thing the field has to a shared data home. (checked 2026-07)
• EXIOBASE: global environmentally-extended MRIO database; free with registration. (checked 2026-07)
• UNEP IRP Global Material Flows Database: country-level material extraction, trade, and consumption. Free. (checked 2026-07)
• ecoinvent: the standard LCI database. Paid license; check what your institution already has. (checked 2026-07)
• RELab: open platform for collecting product-disassembly data, developed at CML, Leiden University. (checked 2026-07)

General ML dataset sources:

• Hugging Face Datasets: the default host for ML datasets; one load_dataset() call away in Python. (checked 2026-07)
• OpenML: curated, uniformly formatted datasets with tasks and benchmark results attached; strong sklearn/mlr integration. (checked 2026-07)
• Kaggle: broad dataset collection plus competitions; quality varies, licenses matter. (checked 2026-07)
• Zenodo: DOI-issuing research repository; where to publish your own datasets as much as find others’. (checked 2026-07)
• UCI ML Repository: the classic benchmark collection; small, clean, good for baselines. (checked 2026-07)

Pretrained models

Where to find models you can download and run (locally or otherwise):

• Hugging Face Hub: the central registry for pretrained models (LLMs, CV, audio); filter by task, license, and size. Most models on the applications map live here. (checked 2026-07)
• Ollama library: curated open-weight LLMs packaged for one-command local use; pairs with the local inference servers. (checked 2026-07)
• Kaggle Models: model registry tied to Kaggle’s datasets and notebooks. (checked 2026-07)
• timm: the standard collection of pretrained vision backbones for PyTorch. (checked 2026-07)

Check the license before building on a model: “open weights” ranges from Apache-2.0 to research-only.