Files
Kepler-8B-Instruct-v2/mergekit_config.yml

15 lines
305 B
YAML
Raw Permalink Normal View History

slices:
- sources:
- model: Qwen/Qwen2.5-7B-Instruct
layer_range: [0, 28]
- model: deepseek-ai/DeepSeek-R1-Distill-Qwen-7B
layer_range: [0, 28]
merge_method: slerp
base_model: Qwen/Qwen2.5-7B-Instruct
parameters:
t:
- value: 0.5
dtype: bfloat16
tokenizer_source: base