Gemma2-Eclipse-9B
The merge script was taken from MarinaraSpaghetti/NemoMix-Unleashed-12B. All I did was replace the models (and I couldn't find a replacement for one of the models so I just increased the weight on one) and it seems to work pretty well.
ChatML prompt format seems to work fine. Gemma 2 format seems to also work.
ChatML seems cause it to incorrectly spew this at the end |im_end|> so Gemma 2 format might work better.
This is a merge of pre-trained language models created using mergekit.
Merge Details
Merge Method
This model was merged using the della_linear merge method using unsloth/gemma-2-9b as a base.
Models Merged
The following models were included in the merge:
- crestf411/gemma2-9B-sunfall-v0.5.2
- anthracite-org/magnum-v4-9b
- unsloth/gemma-2-9b-it
- nbeerbower/Gemma2-Gutenberg-Doppel-9B
Configuration
The following YAML configuration was used to produce this model:
models:
- model: unsloth/gemma-2-9b-it
parameters:
weight: 0.22
density: 0.5
- model: nbeerbower/Gemma2-Gutenberg-Doppel-9B
parameters:
weight: 0.2
density: 0.6
- model: crestf411/gemma2-9B-sunfall-v0.5.2
parameters:
weight: 0.25
density: 0.7
- model: anthracite-org/magnum-v4-9b
parameters:
weight: 0.33
density: 0.8
merge_method: della_linear
base_model: unsloth/gemma-2-9b
parameters:
epsilon: 0.05
lambda: 1
dtype: bfloat16
tokenizer_source: base