README.md

DistilGPT2 English language model fine-tuned on mathematical proofs extracted from arXiv.org LaTeX sources from 1992 to 2020.

Proofs have been cleaned up a bit. In particular, they use

• CITE for any citation
• REF for any reference
• MATH for any LaTeX mathematical formula
• CASE: for any \\item or labeled subcase.

For text generation, I recommend prompts such as:

• Let MATH be given.
• By the inductive hypothesis,
• If MATH is a nonempty