Language Models are Super Mario: Absorbing Abilities from Homologous Models as a Free Lunch
Paper • 2311.03099 • Published • 32
It sucks. Alas.
This model was merged using the DARE TIES merge method using mergekit-community/L3.3-Oddbase-70B as a base.
The following models were included in the merge:
The following YAML configuration was used to produce this model:
models:
- model: mergekit-community/ignore_L3.x-Monk-70B
parameters:
density: 0.7
weight: 0.5
- model: mergekit-community/ignore_L3.x-Hero-70B
parameters:
density: 0.9
weight: 1
- model: mergekit-community/ignore_L3.1-Deity-70B
parameters:
density: 0.5
weight: 0.7
merge_method: dare_ties
base_model: mergekit-community/L3.3-Oddbase-70B
tokenizer_source: base