HERETICODER-2.5-3B-IT

Model Description

This is an abliterated decensored version of Qwen2.5-Coder-3B-Instruct using Heretic!

  • License: apache-2.0
  • Abliterated by : Cannae-AI
  • Base model : Qwen/Qwen2.5-Coder-3B-Instruct
Model Refusals KL divergence
HERETICODER-2.5-3B-IT 0/100 0.30
HERETICODER-2.5-7B-IT 1/100 0.00
  • Parameters:
    • direction_index = 27.30
    • attn.o_proj.max_weight = 1.44
    • attn.o_proj.max_weight_position = 33.01
    • attn.o_proj.min_weight = 1.00
    • attn.o_proj.min_weight_distance = 14.98
    • mlp.down_proj.max_weight = 1.15
    • mlp.down_proj.max_weight_position = 28.50
    • mlp.down_proj.min_weight = 0.95
    • mlp.down_proj.min_weight_distance = 16.30
Downloads last month
6
Safetensors
Model size
3B params
Tensor type
BF16
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for Cannae-AI/HERETICODER-2.5-3B-IT

Base model

Qwen/Qwen2.5-3B
Finetuned
(63)
this model
Quantizations
2 models

Collection including Cannae-AI/HERETICODER-2.5-3B-IT