Masterjp123's picture
Update README.md
ca2d3f2 verified
---
base_model:
- Riiid/sheep-duck-llama-2-13b
- IkariDev/Athena-v4
- TheBloke/Llama-2-13B-fp16
- KoboldAI/LLaMA2-13B-Psyfighter2
- KoboldAI/LLaMA2-13B-Erebus-v3
- Henk717/echidna-tiefigther-25
- Undi95/Unholy-v2-13B
- ddh0/EstopianOrcaMaid-13b
tags:
- mergekit
- merge
- not-for-all-audiences
- ERP
- RP
- Roleplay
- uncensored
- GPTQ
license: llama2
language:
- en
inference: false
---
# Model
This is the GPTQ 4bit quantized version of SnowyRP
[BF16](https://huggingface.co/Masterjp123/SnowyRP-FinalV1-L2-13B)
[GPTQ](https://huggingface.co/Masterjp123/SnowyRP-FinalV1-L2-13B-GPTQ)
[GGUF](https://huggingface.co/Masterjp123/SnowyRP-FinalV1-L2-13B-GGUF)
Any Future Quantizations I am made aware of will be added.
## Merge Details
just used highly ranked modles to try and get a better result, Also I made sure that Model incest would not be a BIG problem by merging models that are pretty pure.
These models CAN and WILL produce X rated or harmful content, due to being heavily uncensored in a attempt to not limit or make the model worse.
This Model has a Very good knowledge base and understands anatomy decently, Plus this Model is VERY versitle and is great for General assistant work, RP and ERP, RPG RPs and much more.
## Model Use:
This model is very good... WITH THE RIGHT SETTINGS.
I personally use microstat mixed with dynamic temp with epsion cut off and eta cut off.
```
Optimal Settings (so far)
Microstat Mode: 2
tau: 2.95
eta: 0.05
Dynamic Temp
min: 0.25
max: 1.8
Cut offs
epsilon: 3
eta: 3
```
Go to the [BF16](https://huggingface.co/Masterjp123/SnowyRP-FinalV1-L2-13B) Repo for more usage settings.
### Merge Method
This model was merged using the [ties](https://arxiv.org/abs/2306.01708) merge method using [TheBloke/Llama-2-13B-fp16](https://huggingface.co/TheBloke/Llama-2-13B-fp16) as a base.
### Models Merged
The following models were included in the merge:
* [Riiid/sheep-duck-llama-2-13b](https://huggingface.co/Riiid/sheep-duck-llama-2-13b)
* [IkariDev/Athena-v4](https://huggingface.co/IkariDev/Athena-v4)
* [KoboldAI/LLaMA2-13B-Psyfighter2](https://huggingface.co/KoboldAI/LLaMA2-13B-Psyfighter2)
* [KoboldAI/LLaMA2-13B-Erebus-v3](https://huggingface.co/KoboldAI/LLaMA2-13B-Erebus-v3)
* [Henk717/echidna-tiefigther-25](https://huggingface.co/Henk717/echidna-tiefigther-25)
* [Undi95/Unholy-v2-13B](https://huggingface.co/Undi95/Unholy-v2-13B)
* [EstopianOrcaMaid](https://huggingface.co/ddh0/EstopianOrcaMaid-13b)
### Configuration
The following YAML configuration was used to produce this model:
for P1
```yaml
base_model:
model:
path: TheBloke/Llama-2-13B-fp16
dtype: bfloat16
merge_method: task_arithmetic
slices:
- sources:
- layer_range: [0, 40]
model:
model:
path: TheBloke/Llama-2-13B-fp16
- layer_range: [0, 40]
model:
model:
path: Undi95/Unholy-v2-13B
parameters:
weight: 1.0
- layer_range: [0, 40]
model:
model:
path: Henk717/echidna-tiefigther-25
parameters:
weight: 0.45
- layer_range: [0, 40]
model:
model:
path: KoboldAI/LLaMA2-13B-Erebus-v3
parameters:
weight: 0.33
```
for P2
```yaml
base_model:
model:
path: TheBloke/Llama-2-13B-fp16
dtype: bfloat16
merge_method: task_arithmetic
slices:
- sources:
- layer_range: [0, 40]
model:
model:
path: TheBloke/Llama-2-13B-fp16
- layer_range: [0, 40]
model:
model:
path: KoboldAI/LLaMA2-13B-Psyfighter2
parameters:
weight: 1.0
- layer_range: [0, 40]
model:
model:
path: Riiid/sheep-duck-llama-2-13b
parameters:
weight: 0.45
- layer_range: [0, 40]
model:
model:
path: IkariDev/Athena-v4
parameters:
weight: 0.33
```
for the final merge
```yaml
base_model:
model:
path: TheBloke/Llama-2-13B-fp16
dtype: bfloat16
merge_method: ties
parameters:
int8_mask: 1.0
normalize: 1.0
slices:
- sources:
- layer_range: [0, 40]
model:
model:
path: ddh0/EstopianOrcaMaid-13b
parameters:
density: [1.0, 0.7, 0.1]
weight: 1.0
- layer_range: [0, 40]
model:
model:
path: Masterjp123/snowyrpp1
parameters:
density: 0.5
weight: [0.0, 0.3, 0.7, 1.0]
- layer_range: [0, 40]
model:
model:
path: Masterjp123/snowyrpp2
parameters:
density: 0.33
weight:
- filter: mlp
value: 0.5
- value: 0.0
- layer_range: [0, 40]
model:
model:
path: TheBloke/Llama-2-13B-fp16
```