Language Models are Super Mario: Absorbing Abilities from Homologous Models as a Free Lunch
Paper • 2311.03099 • Published • 33
Unleash her demons... Merged the best roleplay model with the best uncensored model to date. The outputs are quite good and verbose.
This is a merge of pre-trained language models created using mergekit.
This model was merged using the DARE TIES merge method using Sao10K/L3-8B-Stheno-v3.2 as a base.
The following models were included in the merge:
The following YAML configuration was used to produce this model:
models:
- model: /root/progs/auto-ollama/scripts/Daredevil-8B-abliterated
parameters:
density: [1, 0.7, 0.1] # density gradient
weight: 1.0
merge_method: dare_ties
base_model: /root/progs/auto-ollama/scripts/L3-8B-Stheno-v3.2
parameters:
normalize: true
int8_mask: true
dtype: float16