Language Models are Super Mario: Absorbing Abilities from Homologous Models as a Free Lunch
Paper
β’
2311.03099
β’
Published
β’
30
This is a merge of pre-trained language models created using mergekit.
This model was merged using the DARE TIES merge method using Muhammad2003/Llama3-8B-OpenHermes-DPO as a base.
The following models were included in the merge:
The following YAML configuration was used to produce this model:
models:
- model: Muhammad2003/Llama3-8B-OpenHermes-DPO
- model: birgermoell/llama-3-merge-disco-neural-pace
parameters:
density: 0.53
weight: 0.6
merge_method: dare_ties
base_model: Muhammad2003/Llama3-8B-OpenHermes-DPO
parameters:
int8_mask: true
dtype: bfloat16