Language Models are Super Mario: Absorbing Abilities from Homologous Models as a Free Lunch
Paper
•
2311.03099
•
Published
•
30
Gilded-Tempest-12B is a merge of pre-trained language models created using mergekit.
This model was merged using the DARE TIES merge method using elinas/Chronos-Gold-12B-1.0 as a base.
The following models were included in the merge:
The following YAML configuration was used to produce this model:
merge_method: dare_ties
base_model: elinas/Chronos-Gold-12B-1.0
models:
- model: elinas/Chronos-Gold-12B-1.0
parameters:
weight: 0.333
density: 0.5
- model: Nitral-AI/Captain-Eris_Violet-V0.420-12B
parameters:
weight: 0.333
density: 0.5
- model: FallenMerick/MN-Violet-Lotus-12B
parameters:
weight: 0.333
density: 0.5
parameters:
int8_mask: true
normalize: true
consistency_tolerance: 0.05
anneal_weights: true
enable_permutation: true
dtype: bfloat16