Resolving Interference When Merging Models
Paper • 2306.01708 • Published • 19
How to use yamatazen/SnowElf-12B-v2 with Transformers:
# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("yamatazen/SnowElf-12B-v2", dtype="auto")This is a merge of pre-trained language models created using mergekit.
This model was merged using the TIES merge method using yamatazen/HMS-Slerp-12B-v2 as a base.
The following models were included in the merge:
The following YAML configuration was used to produce this model:
base_model: yamatazen/HMS-Slerp-12B-v2
models:
- model: yamatazen/SnowElf-12B
parameters:
density: 0.6
weight: 0.6
- model: yamatazen/BlueLight-12B
parameters:
density: 0.5
weight: 0.3
merge_method: ties
dtype: bfloat16
parameters:
normalize: true
tokenizer:
source: union