---
base_model:
- SanjiWatsuki/Kunoichi-DPO-v2-7B
- Epiculous/Fett-uccine-Long-Noodle-7B-120k-Context
library_name: transformers
tags:
- mistral
- quantized
- text-generation-inference
- merge
- mergekit
pipeline_tag: text-generation
inference: false

---

> [!TIP]
> **Support:** <br>
> My upload speeds have been cooked and unstable lately. <br>
> Realistically I'd need to move to get a better provider. <br>
> If you **want** and you are able to... <br>
> [**You can support my various endeavors here (Ko-fi).**](https://ko-fi.com/Lewdiculous) <br>
> I apologize for disrupting your experience.


# **GGUF-Imatrix quantizations for [Kunocchini-7b-128k-test](https://huggingface.co./Test157t/Kunocchini-7b-128k-test/).**

# UPDATED: Please download the v2 files that are now available. The new IQ4_NL and IQ4_XS quants were also added.

# What does "Imatrix" mean?

It stands for **Importance Matrix**, a technique used to improve the quality of quantized models.

The **Imatrix** is calculated based on calibration data, and it helps determine the importance of different model activations during the quantization process. The idea is to preserve the most important information during quantization, which can help reduce the loss of model performance.

One of the benefits of using an Imatrix is that it can lead to better model performance, especially when the calibration data is diverse.

More information: [[1]](https://github.com/ggerganov/llama.cpp/discussions/5006) [[2]](https://github.com/ggerganov/llama.cpp/discussions/5263#discussioncomment-8395384)

## *This has been my personal favourite and daily-driver role-play model for a while, so I decided to make new quantizations for it using the full F16-Imatrix data.*

SillyTavern preset files are located [here](https://huggingface.co./Test157t/Kunocchini-7b-128k-test/tree/main/ST%20presets).

*If you want any specific quantization to be added, feel free to ask.*

All credits belong to the [creator](https://huggingface.co./Test157t/).

`Base⇢ GGUF(F16)⇢ GGUF(Quants)`

The new **IQ3_S** merged today has shown to be better than the old Q3_K_S, so I added that instead of the later. Only supported in `koboldcpp-1.59.1` or higher.

Using [llama.cpp](https://github.com/ggerganov/llama.cpp/)-[b2254](https://github.com/ggerganov/llama.cpp/releases/tag/b2254).

For --imatrix data, `imatrix-Kunocchini-7b-128k-test-F16.dat` was used.

# Original model information:

Thanks to @Epiculous for the dope model/ help with llm backends and support overall.

Id like to also thank @kalomaze for the dope sampler additions to ST. 

@SanjiWatsuki Thank you very much for the help, and the model!

ST users can find the TextGenPreset in the folder labeled so.

![image/jpeg](https://cdn-uploads.huggingface.co/production/uploads/642265bc01c62c1e4102dc36/9obNSalcJqCilQwr_4ssM.jpeg)

The following models were included in the merge:
* [SanjiWatsuki/Kunoichi-DPO-v2-7B](https://huggingface.co./SanjiWatsuki/Kunoichi-DPO-v2-7B)
* [Epiculous/Fett-uccine-Long-Noodle-7B-120k-Context](https://huggingface.co./Epiculous/Fett-uccine-Long-Noodle-7B-120k-Context)

### Configuration

The following YAML configuration was used to produce this model:

```yaml
slices:
  - sources:
      - model: SanjiWatsuki/Kunoichi-DPO-v2-7B
        layer_range: [0, 32]
      - model: Epiculous/Fett-uccine-Long-Noodle-7B-120k-Context
        layer_range: [0, 32]
merge_method: slerp
base_model: SanjiWatsuki/Kunoichi-DPO-v2-7B
parameters:
  t:
    - filter: self_attn
      value: [0, 0.5, 0.3, 0.7, 1]
    - filter: mlp
      value: [1, 0.5, 0.7, 0.3, 0]
    - value: 0.5
dtype: bfloat16
```