aeon37/Llama-3.3-8B-Instruct-128K-heretic

This is a decensored version of shb777/Llama-3.3-8B-Instruct-128K, made using Heretic v1.1.0

Abliteration parameters

Parameter	Value
direction_index	15.28
attn.o_proj.max_weight	1.45
attn.o_proj.max_weight_position	21.79
attn.o_proj.min_weight	1.21
attn.o_proj.min_weight_distance	11.77
mlp.down_proj.max_weight	1.22
mlp.down_proj.max_weight_position	29.78
mlp.down_proj.min_weight	0.78
mlp.down_proj.min_weight_distance	17.51

Performance

Metric	This model	Original model (shb777/Llama-3.3-8B-Instruct-128K)
KL divergence	0.0430	0 (by definition)
Refusals	8/100	95/100

---

Llama 3.3 8B 128K Instruct (Fixed)

Original allura-forge/Llama-3.3-8B-Instruct, Thanks!

imatrix GGUF's by mradermacher (Recommended)

static GGUF's

Evals

Additional Fixes:

• Added rope_scaling
• Added chat template (Unsloth) in tokenizer config
• Updated generation config
• Enabled full context length