Merge pull request #1 from errrajaraman/devin/1775747061-add-graders
Rajaraman R+2: Devin AI, rajarr
Unverified signature
a3f5521
* feat: add grader classes for all 3 DRL tasks
- Implement BasicSafetyGrader, ContextNuanceGrader, and FairnessBiasGrader
as proper classes extending openenv Rubric base class
- Add grader registry with get_grader() and list_graders() helpers
- Add grader_id field to TaskConfig and wire each task to its grader
- Update environment.py to use task-specific graders for episode scoring
- Update app.py /graders and /tasks endpoints to use real grader registry
- Update openenv.yaml with entry_point paths for all 3 graders
- Return grader_id in step/predict_and_step responses when episode ends
Co-Authored-By: Rajaraman Ramachandran <errrajaraman@gmail.com>
* fix: include context_type and policy_mode in episode history entries
Ensures ContextNuanceGrader can apply context-aware bonuses/penalties
for quote posts and heated replies instead of always defaulting to
ROOT_POST.
Co-Authored-By: Rajaraman Ramachandran <errrajaraman@gmail.com>
---------
Co-authored-by: Devin AI <158243242+devi…
fix: include context_type and policy_mode in episode history entries
Devin AI+1: rajarr
402a8c7
Ensures ContextNuanceGrader can apply context-aware bonuses/penalties
for quote posts and heated replies instead of always defaulting to
ROOT_POST.
Co-Authored-By: Rajaraman Ramachandran <errrajaraman@gmail.com>
feat: add grader classes for all 3 DRL tasks
Devin AI+1: rajarr
40f1fb9
- Implement BasicSafetyGrader, ContextNuanceGrader, and FairnessBiasGrader
as proper classes extending openenv Rubric base class
- Add grader registry with get_grader() and list_graders() helpers
- Add grader_id field to TaskConfig and wire each task to its grader
- Update environment.py to use task-specific graders for episode scoring
- Update app.py /graders and /tasks endpoints to use real grader registry
- Update openenv.yaml with entry_point paths for all 3 graders
- Return grader_id in step/predict_and_step responses when episode ends
Co-Authored-By: Rajaraman Ramachandran <errrajaraman@gmail.com>
feat: implement state interface, standardize schema IDs, and add task discovery endpoint
Gamucopia-Creatives
a447d83
fix: set default value for ResetRequest in reset_env endpoint to prevent validation errors
Gamucopia-Creatives
46d41dd
feat: implement reinforced human memory for inference and enhance UI with audit capabilities and task labeling
Gamucopia-Creatives
85218c9
refactor: normalize reward range to [0.01, 0.99] and standardize episode scoring key to score across environment and inference logic
Gamucopia-Creatives
da09194
feat: enhance safety moderation with keyword-based fallback, robust LLM response parsing, and unified UI insight formatting.
Gamucopia-Creatives
8ddb260
refactor: update submission validator with strict log format checks, HF_TOKEN safety verification, and add FastAPI server for interactive task testing
Gamucopia-Creatives
1191a4e
UI: Move RUN MODERATION button above PURGE LOGS button
