metadata
language:
- en
library_name: transformers
tags:
- gpt
- llm
- large language model
inference: true
license: other
This model is a merge of euclaise/Echo-3B, stabilityai/stablelm-zephyr-3b and Aryanne/Astridboros-3B using task_arithmetic(see astrea-rp-v1-3b.yml or below).
merge_method: task_arithmetic
base_model: euclaise/Ferret-3B
models:
- model: euclaise/Ferret-3B
- model: stabilityai/stablelm-zephyr-3b
parameters:
weight: 0.33
- model: euclaise/Echo-3B
parameters:
weight: 0.66
- model: Aryanne/Astridboros-3B
parameters:
weight: 0.16
dtype: float16
I recommend the use of Vicuna prompt format, but it's your choice to see what works for you.
I think zephyr license applies to this merge, for non commercial use.