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 /workspace/prototype-0.4x219 as a base.
The following models were included in the merge:
The following YAML configuration was used to produce this model:
models:
- model: /workspace/cache/models--TheDrummer--Anubis-70B-v1.1/snapshots/0a82172eda5b6344505254fc73eea0b444f348ff
parameters:
weight: 0.2
density: 0.7
epsilon: 0.2
lambda: 1.1
- model: /workspace/cache/models--TheDrummer--Fallen-Llama-3.3-70B-v1/snapshots/d46ef2629f1c3cd46789a55793c5ff0af60de3e8
parameters:
weight: 0.2
density: 0.7
epsilon: 0.2
lambda: 1.1
- model: /workspace/cache/models--ReadyArt--L3.3-The-Omega-Directive-70B-Unslop-v2.0/snapshots/6c0508d98e6818c191fb376c6b59a2b742e8ab50
parameters:
weight: 0.2
density: 0.7
epsilon: 0.2
lambda: 1.1
- model: /workspace/cache/models--ReadyArt--The-Omega-Directive-L-70B-v1.0/snapshots/c86903b31447ae943b79c4fe8a5147aa5f9b0aee
parameters:
weight: 0.2
density: 0.7
epsilon: 0.2
lambda: 1.1
- model: /workspace/prototype-0.4x219
parameters:
weight: 0.2
density: 0.7
epsilon: 0.1
lambda: 1.0
base_model: /workspace/prototype-0.4x219
merge_method: della_linear
tokenizer:
source: base
pad_to_multiple_of: 8
int8_mask: true
dtype: bfloat16