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 Zyphra/ZR1-1.5B as a base.
The following models were included in the merge:
The following YAML configuration was used to produce this model:
models:
- model: Zyphra/ZR1-1.5B
parameters:
weight: 5
- model: nvidia/AceMath-1.5B-Instruct
parameters:
weight: 4
- model: Spestly/AwA-1.5B
parameters:
weight: 3
- model: MadeAgents/Hammer2.1-1.5b
parameters:
weight: 3
- model: SakanaAI/TinySwallow-1.5B-Instruct
parameters:
weight: 2
- model: Amu/t1-1.5B
parameters:
weight: 2
tokenizer:
source: union
merge_method: dare_ties
base_model: Zyphra/ZR1-1.5B
dtype: bfloat16