Skip to content

[AutoRL-Bench] Fix ALFWorld prompt exposure, update DeepSearchQA split, and translate tracked files to English#1367

Closed
couragec wants to merge 5 commits intomicrosoft:mainfrom
couragec:pr/autorl-bench-upstream-sync
Closed

[AutoRL-Bench] Fix ALFWorld prompt exposure, update DeepSearchQA split, and translate tracked files to English#1367
couragec wants to merge 5 commits intomicrosoft:mainfrom
couragec:pr/autorl-bench-upstream-sync

Conversation

@couragec
Copy link
Collaborator

@couragec couragec commented Mar 19, 2026

Summary

This PR updates AutoRL-Bench in three ways:

  1. Stop exposing react_prompts.json to agents in ALFWorld.
  2. Update DeepSearchQA to use the intended local split protocol.
  3. Translate tracked AutoRL-Bench files and documentation into English for open-source readiness.

Changes

  • ALFWorld

    • remove task-specific prompt exposure from the mounted workspace
  • DeepSearchQA

    • switch to deterministic local split
    • update the default protocol to 100 train / 200 eval
  • Documentation / open-source cleanup

    • translate tracked files under rdagent/scenarios/rl/autorl_bench
    • add bilingual README support

Notes

  • This branch is based on the latest upstream/main.
  • The PR keeps the existing fork-side AutoRL-Bench commits and merges in recent upstream updates.

📚 Documentation preview 📚: https://RDAgent--1367.org.readthedocs.build/en/1367/

@couragec couragec closed this Mar 19, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant