---
license: other
---
All datasets used for the patterns project.

Language datasets (c4, Fineweb, Open-Web-Math, and Codeparrot) are tokenized with the [HuggingFaceTB/SmolLM2-135M](https://huggingface.co/HuggingFaceTB/SmolLM2-135M) tokenizer.

See runs in https://wandb.ai/bonn/Patterns/workspace?nw=nwusernklugecorrea.