This is a merge of pre-trained language models created using
mergekit.
This model was merged using the
SLERP merge method.
1
2
3# Umbrella Corporation Official Merge Protocol v3.2
4# Author: Dr. Novaciano
5# Objective: Test
6# PROJECT: Progenitor-Virus-3.2-3B
7
8models:
9 - model: DavidAU/Llama-3.2-3B-Instruct-heretic-ablitered-uncensored # Experimental viral strain neural imprint
10 - model: bunnycore/Llama-3.2-3B-Sci-Think # Baseline cognitive template, "safe mode"
11
12merge_method: slerp # Spherical Linear Interpolation to preserve extreme viral traits smoothly
13base_model: DavidAU/Llama-3.2-3B-Instruct-heretic-ablitered-uncensored # Anchor model for stable latent space
14
15dtype: bfloat16 # Memory-efficient precision, minimal loss in viral feature fidelity
16
17parameters:
18 t: 0.45
19 normalize: false
20 rescale: true
21 rescale_factor: 1.12
22 memory_efficient: true
23 low_cpu_mem_usage: true
24
25layer_range:
26 - value: [4, 22]
27
28tie_word_embeddings: true
29tie_output_embeddings: true
30
31
32
33
