Resolving Interference When Merging Models
Paper • 2306.01708 • Published • 18
Slightly different from v4, I shifted the weighting and density towards Celeste slightly more, which seems to have nicer results.
static quants at: https://huggingface.co/mradermacher/MN-12B-Starcannon-v5-unofficial-GGUF
imatrix quants at: https://huggingface.co/mradermacher/MN-12B-Starcannon-v5-unofficial-i1-GGUF
This is a merge of pre-trained language models created using mergekit.
This model was merged using the TIES merge method using nothingiisreal/MN-12B-Celeste-V1.9 as a base.
The following models were included in the merge:
The following YAML configuration was used to produce this model:
models:
- model: anthracite-org/magnum-12b-v2.5-kto
parameters:
density: 0.2
weight: 0.4
- model: nothingiisreal/MN-12B-Celeste-V1.9
parameters:
density: 0.8
weight: 0.6
merge_method: ties
base_model: nothingiisreal/MN-12B-Celeste-V1.9
parameters:
normalize: true
int8_mask: true
dtype: bfloat16