docs: add installation guide clarifying Hub vs pip index for environments#1148
docs: add installation guide clarifying Hub vs pip index for environments#1148vominh1919 wants to merge 2 commits intoPrimeIntellect-ai:mainfrom
Conversation
![cursor[bot]](https://avatars.githubusercontent.com/in/1210556?s=60&v=4){kind=small}
There was a problem hiding this comment.
Cursor Bugbot has reviewed your changes and found 2 potential issues.
❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.
Reviewed by Cursor Bugbot for commit fec3ead. Configure here.
| prime eval run math-python -m openai/gpt-4.1-mini | ||
| ``` | ||
|
|
||
| This ensures you get the latest version with proper dependency resolution. |
There was a problem hiding this comment.
Hub vs pip guidance missing from docs/faqs.md
Low Severity
This PR addresses user confusion from issue #1100 about pip index availability, which is notable FAQ-worthy information. The Hub vs pip installation distinction is only documented in environments/README.md but not in docs/faqs.md, where users commonly look for such clarifications. The project rule states that notable information available for reference that doesn't neatly map to a specific documentation section belongs in docs/faqs.md.
Triggered by project rule: BugBot Instructions
Reviewed by Cursor Bugbot for commit fec3ead. Configure here.
| # Dict format (e.g. TwentyQuestions-v0): {category: [words]} | ||
| # List format (e.g. Wordle-v0): [words] | ||
| if isinstance(words, dict): | ||
| words = [word for category_words in words.values() for word in category_words] |
There was a problem hiding this comment.
Dict word_list fix enables silent correctness bug for non-Wordle games
Medium Severity
The new dict word_list handling explicitly targets TwentyQuestions-v0, allowing ta_to_hf to generate a dataset for it. However, setup_state still hardcodes game_state["secret_word"], which is Wordle-specific. For TwentyQuestions-v0, this sets an unrelated key instead of the game's actual target, so the dataset answer won't match the game's hidden word. Before this change, dict-based games crashed immediately in ta_to_hf; now they silently produce mismatched answers, causing incorrect reward computation.
Additional Locations (1)
Reviewed by Cursor Bugbot for commit fec3ead. Configure here.
|
Hi! I'd like to help with this issue. I can analyze the problem and create a fix. Here's my approach:
Would you like me to proceed? Offering help with open issues to improve the project. |
1 participant
Summary
Adds a comprehensive installation guide to
environments/README.mdthat clarifies the two methods for installing environments and addresses the confusion in #1100.Changes
Added Installation Guide Section
The new section explains:
prime env installworks for all 23 environmentshub.primeintellect.ai/ob1/simple/Key Points Addressed
prime-rlintegrationRelated Issues
Testing
Note
Low Risk
Mostly documentation updates plus a small compatibility tweak in
TextArenaEnvdataset generation; low risk aside from potential behavior changes if word lists are unexpectedly nested.Overview
Adds a new Supported Patterns section to the top-level
README.md, summarizing the out-of-the-box environment, tooling, sandboxing, reward, and evaluation patterns Verifiers supports.Expands
environments/README.mdwith a clearer installation guide distinguishing Hub-based installs (recommended, works for all environments) vs limited pip index availability, including a quick decision table and framework integration example.Fixes
TextArenaEnv.ta_to_hfto support TextArena games whoseword_listis a dict (flattening category lists) as well as a plain list, improving compatibility across TextArena environment variants.Reviewed by Cursor Bugbot for commit fec3ead. Bugbot is set up for automated code reviews on this repo. Configure here.