---
license: apache-2.0
language:
- en
inference: false
tags:
- roleplay
- llama3
- sillytavern
---

# #roleplay #sillytavern #llama3

My GGUF-IQ-Imatrix quants for [**nbeerbower/llama-3-Stheno-Mahou-8B**](https://huggingface.co/nbeerbower/llama-3-Stheno-Mahou-8B).

"A potential precious hidden gem, will you polish this rough diamond?"

This is a merge of two very interesting models, aimed at roleplaying usage.

![image/png](https://cdn-uploads.huggingface.co/production/uploads/65d4cf2693a0a3744a27536c/mhMDEV-VH2cbHvThdZA0T.png)

> [!TIP]
> **Personal-support:** <br>
> I apologize for disrupting your experience. <br>
> Currently I'm working on moving for a better internet provider. <br>
> If you **want** and you are **able to**... <br>
> You can [**spare some change over here (Ko-fi)**](https://ko-fi.com/Lewdiculous). <br>
>
> **Author-support:** <br>
> You can support the author [**at their own page**](https://huggingface.co/nbeerbower).

> [!IMPORTANT]
> **Quantization process:** <br>
> For future reference, these quants have been done after the fixes from [**#6920**](https://github.com/ggerganov/llama.cpp/pull/6920) have been merged. <br>
> Imatrix data was generated from the FP16-GGUF and the final conversions used BF16-GGUF for the quantization process. <br>
> This was a bit more disk and compute intensive but hopefully avoided any losses during conversion. <br>
> If you noticed any issues let me know in the discussions.

> [!NOTE]
> **General usage:** <br>
> Use the latest version of **KoboldCpp**. <br>
> Remember that you can also use `--flashattention` on KoboldCpp now even with non-RTX cards for reduced VRAM usage. <br>
> For **8GB VRAM** GPUs, I recommend the **Q4_K_M-imat** quant for up to 12288 context sizes. <br>
> For **12GB VRAM** GPUs, the **Q5_K_M-imat** quant will give you a great size/quality balance. <br>
>
> **Resources:** <br>
> You can find out more about how each quant stacks up against each other and their types [**here**](https://gist.github.com/Artefact2/b5f810600771265fc1e39442288e8ec9) and [**here**](https://rentry.org/llama-cpp-quants-or-fine-ill-do-it-myself-then-pt-2), respectively.
> 
> **Presets:** <br>
> Some compatible SillyTavern presets can be found [**here (Virt's Roleplay Presets)**](https://huggingface.co/Virt-io/SillyTavern-Presets), experiment with Llama-3 and ChatML. <br>
<!-- > Check [**discussions such as this one**](https://huggingface.co/Virt-io/SillyTavern-Presets/discussions/5#664d6fb87c563d4d95151baa) for other recommendations and samplers.
-->

## **Original model text information:**

# llama-3-Stheno-Mahou-8B

This is a merge of pre-trained language models created using [mergekit](https://github.com/cg123/mergekit).

## Merge Details
### Merge Method

This model was merged using the [Model Stock](https://arxiv.org/abs/2403.19522) merge method using [flammenai/Mahou-1.2-llama3-8B](https://huggingface.co/flammenai/Mahou-1.2-llama3-8B) as a base.

### Models Merged

The following models were included in the merge:
* [flammenai/Mahou-1.1-llama3-8B](https://huggingface.co/flammenai/Mahou-1.1-llama3-8B)
* [Sao10K/L3-8B-Stheno-v3.1](https://huggingface.co/Sao10K/L3-8B-Stheno-v3.1)

### Configuration

The following YAML configuration was used to produce this model:

```yaml
models:
  - model: flammenai/Mahou-1.1-llama3-8B
  - model: Sao10K/L3-8B-Stheno-v3.1
merge_method: model_stock
base_model: flammenai/Mahou-1.2-llama3-8B
dtype: bfloat16


```