UncensoredLLms
Collection
This is an abliterated decensored LLm collection by Cannae Ai!
•
14 items
•
Updated
•
1
GPT-5-Llama-abliterated
Model Description
This is an abliterated decensored version of GPT-5-Distill-llama3.2-3B-Instruct!
- License: apache-2.0
- Abliterated by : Cannae-AI
| Model | Refusals | KL divergence |
|---|---|---|
| GPT-5-Llama-abliterated | 6/100 | 0.03 |
Abliteration Parameters:
- direction_index = 14.09
- attn.o_proj.max_weight = 1.44
- attn.o_proj.max_weight_position = 18.50
- attn.o_proj.min_weight = 1.43
- attn.o_proj.min_weight_distance = 13.36
- mlp.down_proj.max_weight = 0.95
- mlp.down_proj.max_weight_position = 26.42
- mlp.down_proj.min_weight = 0.35
- mlp.down_proj.min_weight_distance = 15.94
- Downloads last month
- 2
Model tree for Cannae-AI/GPT-5-Llama-abliterated
Base model
meta-llama/Llama-3.2-3B-Instruct
Finetuned
unsloth/Llama-3.2-3B-Instruct