Disobedience rate: 9%, original: 97%
KL divergence: 0.5896
Parameters:
direction_index = 13.34
attn.o_proj.max_weight = 1.42
attn.o_proj.max_weight_position = 14.41
attn.o_proj.min_weight = 0.83
attn.o_proj.min_weight_distance = 7.49
mlp.down_proj.max_weight = 1.37
mlp.down_proj.max_weight_position = 14.72
mlp.down_proj.min_weight = 1.06
mlp.down_proj.min_weight_distance = 5.89
- Downloads last month
- 46
Model tree for hereticness/Heretic-Bellatrix-Tiny-1B
Base model
meta-llama/Llama-3.2-1B-Instruct
Finetuned
prithivMLmods/Bellatrix-Tiny-1B