Language Models are Super Mario: Absorbing Abilities from Homologous Models as a Free Lunch
Paper • 2311.03099 • Published • 32
Experimental model reccomended by this user: https://huggingface.co/VivyAI
It turned out quite interesting, I will be iterating on this at least a few times. Thank you for the suggestions VivyAI!
This is a merge of pre-trained language models created using mergekit.
This model was merged using the DARE TIES merge method using jeiku/Rosa_v1_3B as a base.
The following models were included in the merge:
The following YAML configuration was used to produce this model:
models:
- model: RosaHermesBoros+jeiku/No_Robots_Alpaca_StableLM
parameters:
weight: 0.25
density: 0.25
- model: RosaHermesBoros+jeiku/Toxic_DPO_StableLM
parameters:
weight: 0.25
density: 0.25
merge_method: dare_ties
base_model: jeiku/Rosa_v1_3B
parameters:
dtype: bfloat16