Nice work! Interested in the design of 1 vs 1 battles between LVLMs, but can you share more details about the Elo rating algorithm? Like the choice of k-factor, the expected confidence intervals with the collected user ratings, etc. Appreciated if you can share more of the details.
Nice work! Interested in the design of 1 vs 1 battles between LVLMs, but can you share more details about the Elo rating algorithm? Like the choice of k-factor, the expected confidence intervals with the collected user ratings, etc. Appreciated if you can share more of the details.