Disobedience rate: 7%, original: 71%
KL divergence: 0.0577
Parameters:
direction_index = 16.81
attn.o_proj.max_weight = 1.48
attn.o_proj.max_weight_position = 19.55
attn.o_proj.min_weight = 1.31
attn.o_proj.min_weight_distance = 13.65
mlp.down_proj.max_weight = 0.82
mlp.down_proj.max_weight_position = 22.91
mlp.down_proj.min_weight = 0.48
mlp.down_proj.min_weight_distance = 3.16
- Downloads last month
- 1