FuseChat: Knowledge Fusion of Chat Models
Paper
•
2408.07990
•
Published
•
14
This is a merge of pre-trained language models created using mergekit.
This model was merged using the SCE merge method using CultriX/Qwen2.5-14B-Wernicke as a base.
The following models were included in the merge:
The following YAML configuration was used to produce this model:
models:
- model: sometimesanotion/Lamarck-14B-v0.7
parameters:
select_topk: 1.0
- model: arcee-ai/SuperNova-Medius
parameters:
select_topk: 0.003125
- model: sometimesanotion/Qwen2.5-14B-Vimarckoso-v3
parameters:
select_topk: 0.00625
- model: deepseek-ai/DeepSeek-R1-Distill-Qwen-14B
parameters:
select_topk: 0.003125
- model: sometimesanotion/Qwenvergence-14B-v10-DS
parameters:
select_topk: 0.275
merge_method: sce
base_model: CultriX/Qwen2.5-14B-Wernicke
tokenizer:
source: sometimesanotion/Lamarck-14B-v0.7
dtype: float32
out_dtype: bfloat16