Key Info

OpenAI's math problems have been turned into open-source RL environments on Hugging Face. The release is described as early and rough, with the verifier only accepting the exact original formalization, so an equivalent proof can still score 0.

Highlights

  • OpenAI math problems converted into executable open-source RL environments
  • Verifier only accepts the exact original formalization; equivalent proofs may score 0
  • Framed as turning open research into executable training infrastructure for everyone
  • Aimed at letting anyone build on it to train better open models