🏆 The GPU-Poor LLM Gladiator Arena 🏆 v25.03

Step right up to the arena where frugal meets fabulous in the world of AI! Watch as our compact contenders (maxing out at 14B parameters) duke it out in a battle of wits and words.

What started as a simple experiment has grown into a popular platform for evaluating compact language models. As the arena continues to expand with more models, features, and battles, it requires computational resources to maintain and improve. If you find this project valuable and would like to support its development, consider sponsoring:
  1. To start the battle, go to the 'Battle Arena' tab.
  2. Type your prompt into the text box. Alternatively, click the "🎲" button to receive a random prompt.
  3. Click the "Generate Responses" button to view the models' responses.
  4. Cast your vote for the model that provided the better response. In the event of a Tie, enter a new prompt before continuing the battle.
  5. Check out the Leaderboard to see how models rank against each other.

More info: README.md

Main Leaderboard

This leaderboard uses a scoring system that balances win rate and total battles. The score is calculated using the formula: Score = Win Rate * (1 - 1 / (Total Battles + 1))

This formula rewards models with higher win rates and more battles. As the number of battles increases, the score approaches the win rate.

Leaderboard

Leaderboard