Model soups: averaging weights of multiple fine-tuned models improves accuracy without increasing inference time
Paper • 2203.05482 • Published • 8
This is a merge of pre-trained language models created using mergekit.
This model was merged using the Linear merge method.
The following models were included in the merge:
The following YAML configuration was used to produce this model:
# Linear Merge: French-Qwen + LUTH (50% interpolation)
# Simple weighted average of model weights
merge_method: linear
dtype: bfloat16
models:
- model: /workspace/french-qwen-asr/models/french-qwen-100k-avg
parameters:
weight: 0.5
- model: /workspace/french-qwen-asr/models/luth-1.7b-instruct
parameters:
weight: 0.5