Language Models are Super Mario: Absorbing Abilities from Homologous Models as a Free Lunch
Paper
•
2311.03099
•
Published
•
30
This is a merge of pre-trained language models created using mergekit.
This model was merged using the DARE TIES merge method using lightblue/suzume-llama-3-8B-japanese as a base.
The following models were included in the merge:
The following YAML configuration was used to produce this model:
models:
- model: lightblue/suzume-llama-3-8B-japanese
# no parameters necessary for base model
- model: ContactDoctor/Bio-Medical-Llama-3-8B
parameters:
density: 0.5
weight: 0.5
merge_method: dare_ties
base_model: lightblue/suzume-llama-3-8B-japanese
parameters:
normalize: true
dtype: float16