DELLA-Merging: Reducing Interference in Model Merging through Magnitude-Based Sampling
Paper
•
2406.11617
•
Published
•
8
This is a merge of pre-trained language models created using mergekit.
This model was merged using the Linear DELLA merge method using Vortex5/MegaMoon-Karcher-12B as a base.
The following models were included in the merge:
The following YAML configuration was used to produce this model:
models:
- model: Vortex5/Vermilion-Sage-12B
parameters:
weight: [0.2, 0.5, 0.9, 1.0, 0.95, 0.8, 0.4, 0.2]
density: 0.55
epsilon: 0.4
- model: Vortex5/Dark-Quill-12B
parameters:
weight: [0.8, 1.0, 0.9, 0.7, 0.5, 0.4, 0.2, 0.0]
density: 0.5
epsilon: 0.4
merge_method: della_linear
base_model: Vortex5/MegaMoon-Karcher-12B
parameters:
lambda: 0.94
normalize: true
dtype: bfloat16
tokenizer:
source: union