Model Breadcrumbs: Scaling Multi-Task Model Merging with Sparse Masks
Paper
• 2312.06795 • Published
• 1
This is a merge of pre-trained language models created using mergekit.
This model was merged using the Model Breadcrumbs merge method using meta-llama/Llama-3.2-1B as a base.
The following models were included in the merge:
The following YAML configuration was used to produce this model:
base_model: meta-llama/Llama-3.2-1B
dtype: float16
merge_method: breadcrumbs
modules:
default:
slices:
- sources:
- layer_range: [0, 16]
model: Alelcv27/llama3-1b-code-dpo
parameters:
weight: 0.9
- layer_range: [0, 16]
model: Alelcv27/llama3-1b-math-dpo
parameters:
weight: 0.8
- layer_range: [0, 16]
model: meta-llama/Llama-3.2-1B
parameters:
density: 0.7
gamma: 0.05
normalize: 1.0