Language Models are Super Mario: Absorbing Abilities from Homologous Models as a Free Lunch
Paper • 2311.03099 • Published • 32
This is a merge of pre-trained language models created using mergekit.
This model was merged using the DARE TIES merge method using KidIkaros/Llama-3.2-1B-Instruct-abliterated as a base.
The following models were included in the merge:
The following YAML configuration was used to produce this model:
slices:
- sources:
- model: KidIkaros/Llama-3.2-1B-Instruct-abliterated
layer_range: [0, 16]
- model: passing2961/Thanos-1B
layer_range: [0, 16]
- model: DeepAutoAI/Explore_Llama-3.2-1B-Inst_v1.1
layer_range: [0, 16]
merge_method: dare_ties
base_model: KidIkaros/Llama-3.2-1B-Instruct-abliterated
parameters:
density: 0.5
weight: 0.4
dtype: bfloat16