DELLA-Merging: Reducing Interference in Model Merging through Magnitude-Based Sampling
Paper • 2406.11617 • Published • 10
This is a merge of pre-trained language models created using mergekit.
This model was merged using the DELLA merge method using arcee-ai/Virtuoso-Medium-v2 as a base.
The following models were included in the merge:
The following YAML configuration was used to produce this model:
models:
- model: arcee-ai/Virtuoso-Medium-v2
- model: nbeerbower/Rombos-EVAGutenberg-TIES-Qwen2.5-32B
parameters:
density: 0.65
weight: 0.35
- model: bespokelabs/Bespoke-Stratos-32B
parameters:
density: 0.5
weight: 0.25
- model: crestf411/Q2.5-32B-Slush
parameters:
density: 0.5
weight: 0.25
merge_method: della
base_model: arcee-ai/Virtuoso-Medium-v2
parameters:
density: 0.55
epsilon: 0.1
lambda: 1.0
normalize: true
int8_mask: true
dtype: bfloat16
tokenizer_source: base