Language Models are Super Mario: Absorbing Abilities from Homologous Models as a Free Lunch
Paper • 2311.03099 • Published • 30
This is quantized version of BarBarickoza/Gemma-Ataraxy-Dare-9b created using llama.cpp
This is a merge of pre-trained language models created using mergekit.
This model was merged using the DARE TIES merge method using lemon07r/Gemma-2-Ataraxy-9B as a base.
The following models were included in the merge:
The following YAML configuration was used to produce this model:
models:
- model: lemon07r/Gemma-2-Ataraxy-9B
# no parameters necessary for base model
- model: anthracite-org/magnum-v3-9b-customgemma2
parameters:
weight: 0.1
density: 0.15
- model: inflatebot/G2-9B-Blackout-R1
parameters:
weight: 0.2
density: 0.4
merge_method: dare_ties
base_model: lemon07r/Gemma-2-Ataraxy-9B
parameters:
int8_mask: true
dtype: bfloat16
tokenizer_source: anthracite-org/magnum-v3-9b-customgemma2
normalize: false
2-bit
3-bit
4-bit
5-bit
6-bit
8-bit