Lewdiculous/EndlessRP-v3-7B-GGUF-Imatrix

GGUF-Imatrix quantizations for Endevor/EndlessRP-v3-7B.

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] [2].

For --imatrix data, imatrix-EndlessRP-v3-7B-F16.dat was used.

Base⇢ GGUF(F16)⇢ Imatrix-Data(F16)⇢ GGUF(Imatrix-Quants)

Using llama.cpp-b2280.

The new IQ3_S quant-option 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.

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

All credits belong to the creator.

Waifu mascot:

Original model information:

This model is recommended for RP, but you can use it as assistant as well.

New model! Version 2 brings less GPTims, but it's more the same, so I made this one. This is probably the best. Please, give it a try.

---

Prompt Format:

• Extended Alpaca Format As for exemple from lemonilia/LimaRP-Mistral-7B-v0.1. Use ### Response: (length = huge) for exemple, to increase length. You can use Metharme or ChatML as well, but Alpaca is recommended.

Configuration

Source:

models:
  - model: mistralai/Mistral-7B-v0.1
  - model: Elizezen/Hameln-japanese-mistral-7B
    # This model brings very good creative output...
    parameters:
      density: 0.6
      weight: 0.25
  - model: fblgit/una-cybertron-7b-v3-OMA+.\toxic-dpo-v0.1-NoWarning-lora
    # Please, refer to model page for more information. Added a finetuned Toxic DPO to remove some boring warnings. 
    parameters:
      density: 0.6
      weight: 0.25
  - model: cgato/Thespis-CurtainCall-7b-v0.1.2+Doctor-Shotgun/mistral-v0.1-7b-pippa-metharme-lora
    # A good model compartible with ST. I added a PIPPA + METHARME lora to make it more 'balanced'.
    parameters:
      density: 0.6
      weight: 0.25
merge_method: dare_ties
base_model: mistralai/Mistral-7B-v0.1
parameters:
  int8_mask: true
dtype: bfloat16

As this mostly focuses on RP and creating stories, please don't expect it being smart with riddles or logical tests.