[sergiopaniego]

Each GRPO step opens 64 sessions, and openenv/wordle allows 8 (MAX_CONCURRENT_ENVS). With the official Space as the default, a run without --env-url fails partway into the first step. Now grpo.py requires --env-url, so it fails at startup, and the README and tutorial already say to pass the URL of your copy.

evaluate.py keeps its default, since it plays one game at a time. Both scripts now call get_token() once.

Same change as in the TRL Wordle example (huggingface/trl#7932).

---

Note

Low Risk
CLI and test-only changes for the Wordle example scripts; no changes to auth logic beyond avoiding a duplicate token lookup.

Overview
GRPO training now requires --env-url at startup (no default to the public openenv-wordle Space). That avoids a mid-step failure when a run opens many concurrent sessions against a Space capped at 8 environments while GRPO needs 64.

evaluate.py still defaults --env-url for single-game runs. evaluate.py and grpo.py both call get_token() once when building optional auth headers for private Spaces.

Tests pass --env-url when exercising grpo.py.

^{Reviewed by Cursor Bugbot for commit 4c8463d. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes using default effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 3158933. Configure here.}