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 cgato/Nemo-12b-Humanize-SFT-v0.2-Quarter as a base.
The following models were included in the merge:
The following YAML configuration was used to produce this model:
base_model: cgato/Nemo-12b-Humanize-SFT-v0.2-Quarter
config_source: anthracite-org/magnum-v2-12b
tokenizer_source: anthracite-org/magnum-v2-12b
dtype: float16
merge_method: dare_ties
parameters:
normalize: true
int8_mask: true
chat_template: "chatml"
models:
- model: anthracite-org/magnum-v2-12b
parameters:
weight: 0.4
- model: cgato/Nemo-12b-Humanize-SFT-v0.2-Quarter
parameters:
weight: 0.7