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 /workspace/prototype-0.4x197 as a base.
The following models were included in the merge:
The following YAML configuration was used to produce this model:
models:
- model: /workspace/cache/models--bruhzair--prototype-0.4x195/snapshots/a1cb4161a717ebf8052ce09dccaedf2dde2a7a9f
parameters:
weight: 0.15
density: 0.4
- model: /workspace/prototype-0.4x204
parameters:
weight: 0.20
density: 0.4
- model: /workspace/prototype-0.4x208
parameters:
weight: 0.15
density: 0.4
- model: /workspace/prototype-0.4x210
parameters:
weight: 0.15
density: 0.4
- model: /workspace/prototype-0.4x197
parameters:
weight: 0.5
density: 0.4
merge_method: dare_ties
base_model: /workspace/prototype-0.4x197
parameters:
normalize: false
dtype: bfloat16
pad_to_multiple_of: 8
int8_mask: true
tokenizer:
source: base