Add interactive rollout figure to the RL Environment Pipeline recipe

#4
by burtenshaw HF Staff - opened

Summary

Follow up to PR #3. Add a selectable example between task validation and the GRPO section in the RL Environment Pipeline recipe.

Admitted task โ†’ harness rollout โ†’ verifier reward + captured model calls
                                      โ†“
                         Evaluation evidence or online RL sample

Evidence

  • The merged recipe moves from validation to difficulty scoring. The updated recipe shows four representative rollout outcomes in between.
  • Tested all four outcomes in the full guide at 320 px and 728 px. The controls update the reward and training eligibility without horizontal overflow.

Merge Danger

Reversible. The change adds one self-contained section to the recipe HTML.

burtenshaw changed pull request status to merged

Sign up or log in to comment