ToastyPigeon's picture
Update README.md
b778df8 verified
---
base_model: []
tags:
- mergekit
- merge
---
# Psyonic-Cetacean-20B-V2
This is a merge of pre-trained language models created using [mergekit](https://github.com/cg123/mergekit).
GGUF (standard and iMatrix) quants can be found here courtesy of MarsupialAI: https://huggingface.co/MarsupialAI/Psyonic-Cetacean-20b-v2_iMatrix_GGUF
## Merge Details
### Merge Method
This model was merged using the [linear](https://arxiv.org/abs/2203.05482) merge method on two stack-merged models.
The first is [jebcarter/psyonic-cetacean-20B](https://huggingface.co/jebcarter/psyonic-cetacean-20B)
(Orca first, reproduced so I didn't have to download that model on top of the components).
The second is the same recipe with the models reversed.
Since [jebcarter](https://huggingface.co/jebcarter) suggested this recipe, credit goes to him.
### Models Merged
The following models were included in the merge:
* microsoft/Orca-2-13b
* KoboldAI/LLaMA2-13B-Psyfighter2
### Configuration
The following YAML configuration was used to produce this model:
```yaml
models:
- model: microsoft/Orca-2-13b
parameters:
weight: 1.0
merge_method: task_arithmetic
base_model: TheBloke/Llama-2-13B-fp16
dtype: float16
name: FlatOrca2
---
slices:
- sources:
- model: FlatOrca2
layer_range: [0, 16]
- sources:
- model: KoboldAI/LLaMA2-13B-Psyfighter2
layer_range: [8, 24]
- sources:
- model: FlatOrca2
layer_range: [17, 32]
- sources:
- model: KoboldAI/LLaMA2-13B-Psyfighter2
layer_range: [25, 40]
merge_method: passthrough
dtype: float16
name: Psycet
---
slices:
- sources:
- model: KoboldAI/LLaMA2-13B-Psyfighter2
layer_range: [0, 16]
- sources:
- model: FlatOrca2
layer_range: [8, 24]
- sources:
- model: KoboldAI/LLaMA2-13B-Psyfighter2
layer_range: [17, 32]
- sources:
- model: FlatOrca2
layer_range: [25, 40]
merge_method: passthrough
dtype: float16
name: Psycet-Reverse
---
models:
- model: Psycet
parameters:
weight: 0.5
- model: Psycet-Reverse
parameters:
weight: 0.5
merge_method: linear
dtype: float16
```