0xA50C1A1/Llama-3.3-8B-Instruct-128K-SOM-MPOA

This is a decensored version of shb777/Llama-3.3-8B-Instruct-128K, made using Heretic v1.2.0

Abliteration parameters

Parameter	Value
direction_index	15.07
attn.o_proj.max_weights.0	0: 1.40
attn.o_proj.max_weights.1	1: 0.93
attn.o_proj.max_weights.2	2: 1.06
attn.o_proj.max_weights.3	3: 1.45
attn.o_proj.max_weights.4	4: 1.41
attn.o_proj.max_weights.5	5: 1.45
attn.o_proj.max_weights.6	6: 1.25
attn.o_proj.max_weights.7	7: 1.38
attn.o_proj.max_weight_position	28.70
attn.o_proj.min_weights.0	0: 0.71
attn.o_proj.min_weights.1	1: 0.21
attn.o_proj.min_weights.2	2: 0.94
attn.o_proj.min_weights.3	3: 1.36
attn.o_proj.min_weights.4	4: 0.19
attn.o_proj.min_weights.5	5: 1.09
attn.o_proj.min_weights.6	6: 0.87
attn.o_proj.min_weights.7	7: 1.14
attn.o_proj.min_weight_distance	15.91
mlp.down_proj.max_weights.0	0: 1.38
mlp.down_proj.max_weights.1	1: 1.26
mlp.down_proj.max_weights.2	2: 1.10
mlp.down_proj.max_weights.3	3: 0.85
mlp.down_proj.max_weights.4	4: 1.12
mlp.down_proj.max_weights.5	5: 1.31
mlp.down_proj.max_weights.6	6: 1.25
mlp.down_proj.max_weights.7	7: 0.94
mlp.down_proj.max_weight_position	30.75
mlp.down_proj.min_weights.0	0: 0.27
mlp.down_proj.min_weights.1	1: 0.56
mlp.down_proj.min_weights.2	2: 0.77
mlp.down_proj.min_weights.3	3: 0.47
mlp.down_proj.min_weights.4	4: 0.93
mlp.down_proj.min_weights.5	5: 0.67
mlp.down_proj.min_weights.6	6: 1.19
mlp.down_proj.min_weights.7	7: 0.01
mlp.down_proj.min_weight_distance	6.58

Performance

Metric	This model	Original model (shb777/Llama-3.3-8B-Instruct-128K)
KL divergence	0.0563	0 (by definition)
Refusals	3/100	93/100

---

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