🏆 MascuLead: Bias Leaderboard

*GG = Gender Gap
Leaderboard for cover letters that were generated with both neutral and gendered prompts.

# Model Manual annotation Average (↓) GG Masc-Neutral GG Fem-Neutral GG Masc-Gendered GG Fem-Gendered Gender Shift Corpus size (number of letter) Date
1 xglm-2 13.64 1.08 - 7.05 - 32.79 8898 2025-05-29 08:26:21
2 mistral-7b-v0.3 17.87 0.71 - - 7.73 45.18 7636 2025-05-29 08:26:21
3 croissantbase 24.98 - 8.15 9.07 - 57.71 8322 2025-05-29 08:26:21
4 bloom-560m 27.35 15.82 - 1.15 - 65.09 8902 2025-05-29 08:26:21
5 llama-3.2-3b 27.88 33.05 - 10.05 - 40.54 9422 2025-05-29 08:26:21
6 gemma-2-2b 30.27 23.7 - 10.39 - 56.71 9786 2025-05-29 08:26:21
7 gpt2-fr 31.66 12.81 - 21.81 - 60.35 9704 2025-05-29 08:26:21
8 bloom-7b 32.25 11.04 - 19.93 - 65.78 8322 2025-05-29 08:26:21
9 croissant-chat* 33.88 23.89 - 11.44 - 66.32 9862 2025-05-29 08:26:21
10 bloom-3b 36.00 18.95 - 17.23 - 71.82 8792 2025-05-29 08:26:21
11 mistral-7b-instruct-v0.3* 38.52 47.67 - - - 67.53 9552 2025-05-29 08:26:21
12 gemma-2-2b-it* 44.22 57.18 - 28.88 - 46.59 9190 2025-05-29 08:26:21
13 vigogne-2-7b 50.77 69.23 - 18.4 - 64.69 8768 2025-05-29 08:26:21
14 llama-3.2-3b-it* 58.14 65.57 - 25.47 - 83.37 9934 2025-05-29 08:26:21