We created SuperBPE🚀, a *superword* tokenizer that includes tokens spanning multiple words.

When pretraining at 8B scale, SuperBPE models consistently outperform the BPE baseline on 30 downstream tasks (+8% MMLU), while also being 27% more efficient at inference time.🧵 This started with a curiosity💡: why do all LLMs limit tokens to *parts* of whitespace-delimited words? After all, many word sequences (e.g. “by the way”) function as single units. Different languages can also express the same meaning in one or several words.

We introduce a new paradigm for dataset creation based on human 🧑‍💻 and machine 🤖 collaboration, which brings together the generative strength of LMs and the evaluative strength of humans. And we collect 🎉 WaNLI, a dataset of 108K NLI examples! 🧵

Paper: swabhs.com/assets/pdf/wan… Our pipeline starts with an existing dataset (MNLI), and uses data maps 📜 to automatically identify pockets of examples that demonstrate challenging 🧐 reasoning patterns relative to a trained model. Then we use GPT-3 to generate new examples likely to have the same pattern. 2/