🎯 RLHF Alignment Hub
Preference dataset collection, model arena side-by-side evaluation, and fine-tuning export engine
Total Candidates
0
Pending Annotation
0
Annotated Pairings
0
Top Model Elo
—
Judge vs. Votes
—
Target Domain:
INCOMING MODEL PROMPT (boardroom)
Item #--
Loading candidate prompt from queue...
MODEL A (gemini-3.6-flash)
Not judged yet
Response A loading...
MODEL B (gpt-4o)
Not judged yet
Response B loading...
Model Elo Leaderboard
| Rank | Model Name | Elo Rating | Total Battles | Wins | Losses | Ties | Win Rate |
|---|---|---|---|---|---|---|---|
| Loading Elo metrics... | |||||||
The ladder moves on human votes only, from a 1200 start with k=32. A Tie counts as a draw for both models. A Both Bad vote counts as a battle for both but is not a win, loss or tie, and moves no rating — nothing about it says which answer was better.