This is a decensored version of TheDrummer/Big-Tiger-Gemma-27B-v3, made using Heretic v1.1.0

Abliteration parameters

Parameter Value
direction_index 54.60
attn.o_proj.max_weight 1.18
attn.o_proj.max_weight_position 55.35
attn.o_proj.min_weight 0.61
attn.o_proj.min_weight_distance 34.03
mlp.down_proj.max_weight 0.83
mlp.down_proj.max_weight_position 53.27
mlp.down_proj.min_weight 0.50
mlp.down_proj.min_weight_distance 22.81

Performance

Metric This model Original model (TheDrummer/Big-Tiger-Gemma-27B-v3)
KL divergence 0.2434 0 (by definition)
Refusals 5/100 51/100

Join our Discord! https://discord.gg/BeaverAI or our Reddit! https://www.reddit.com/r/BeaverAI/

More than 6000 members strong πŸ’ͺ A hub for users and makers alike!


Big Tiger Gemma 27B v3

image/png

Description

Gemma 3 27B tune that unlocks more capabilities and less positivity! Should be vision capable.

  • More neutral tone, especially when dealing with harder topics.
  • No em-dashes just for the heck of it.
  • Less markdown responses, more paragraphs.
  • Better steerability to harder themes.

Special Thanks

Links

config-v3a

Downloads last month
40
Safetensors
Model size
27B params
Tensor type
BF16
Β·
Inference Providers NEW
This model isn't deployed by any Inference Provider. πŸ™‹ Ask for provider support

Model tree for coder3101/Big-Tiger-Gemma-27B-v3-heretic

Finetuned
(2)
this model

Collection including coder3101/Big-Tiger-Gemma-27B-v3-heretic