This is a decensored version of axolotl-ai-co/Mistral-Small-4-119B-2603-BF16, made using Heretic v1.2.0

Abliteration parameters

Parameter Value
direction_index per layer
attn.o_proj.max_weight 1.49
attn.o_proj.max_weight_position 22.84
attn.o_proj.min_weight 1.49
attn.o_proj.min_weight_distance 20.26

Performance

Metric This model Original model (axolotl-ai-co/Mistral-Small-4-119B-2603-BF16)
KL divergence 0.0167 0 (by definition)
Refusals 27/100 98/100

Mistral Small 4 119B A6B BF16

This is created from https://huggingface.co/mistralai/Mistral-Small-4-119B-2603 descaled to BF16.

Downloads last month
1,483
Safetensors
Model size
119B params
Tensor type
BF16
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for darkc0de/Mistral-Small-4-119B-2603-heretic

Finetuned
(9)
this model