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 mlabonne/Meta-Llama-3.1-8B-Instruct-abliterated as a base.
The following models were included in the merge:
The following YAML configuration was used to produce this model:
models:
- model: mlabonne/Meta-Llama-3.1-8B-Instruct-abliterated
# No parameters necessary for base model
- model: netcat420/MFANNv0.20
parameters:
density: 0.53
weight: 1
- model: netcat420/MFANNv0.19
parameters:
density: 0.53
weight: 1
- model: netcat420/MFANNv0.19.11
parameters:
density: 0.53
weight: 1
- model: netcat420/MFANNv0.20.11
parameters:
density: 0.53
weight: 1
- model: netcat420/MFANNv0.21
parameters:
density: 0.53
weight: 1
merge_method: dare_ties
base_model: mlabonne/Meta-Llama-3.1-8B-Instruct-abliterated
parameters:
int8_mask: true
dtype: bfloat16