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 mistralai/Codestral-22B-v0.1 as a base.
The following models were included in the merge:
The following YAML configuration was used to produce this model:
models:
- model: MrezaPRZ/codestral_high_quality_sft
parameters:
density: 0.5
weight: 0.2
- model: MrezaPRZ/codestral_high_quality_sft_bigquery
parameters:
density: 0.5
weight: 0.2
- model: MrezaPRZ/codestral_high_quality_sft_postgres
parameters:
density: 0.4
weight: 0.2
merge_method: dare_ties
base_model: mistralai/Codestral-22B-v0.1
parameters:
normalize: true
dtype: bfloat16