---
license: cc-by-nc-4.0
language:
- en
---

# WinterGoddess-1.4x-70B-L2 IQ2-GGUF

## Description
IQ2-GGUF quants of [Sao10K/WinterGoddess-1.4x-70B-L2](https://huggingface.co/Sao10K/WinterGoddess-1.4x-70B-L2)

Unlike regular GGUF quants this uses important matrix similar to Quip# to keep the quant from degrading too much even at 2bpw allowing you to run larger models on less powerful machines.

***NOTE:*** Currently you will need experimental branches of Koboldcpp or Ooba for this to work. 
- Nexesenex have compiled Windows binaries [HERE](https://github.com/Nexesenex/kobold.cpp/releases/tag/v1.55.1_b1842)
- [llamacpp_0.2.29 branch](https://github.com/oobabooga/text-generation-webui/tree/llamacpp_0.2.29) of Ooba also works


[More info about IQ2](https://github.com/ggerganov/llama.cpp/pull/4897)



# Models

Models: [IQ2-XS](https://huggingface.co/Kooten/WinterGoddess-1.4x-70B-L2-IQ2-GGUF/blob/main/WinterGoddess-1.4x-70B-L2-IQ2_XS.gguf), [IQ2-XXS](https://huggingface.co/Kooten/WinterGoddess-1.4x-70B-L2-IQ2-GGUF/blob/main/WinterGoddess-1.4x-70B-L2-IQ2_XXS.gguf)

Regular GGUF Quants: [Here](https://huggingface.co/TheBloke/WinterGoddess-1.4x-70B-L2-GGUF)

## Prompt Format

### Alpaca:
```
### Instruction:
<Prompt>

### Response:

```

OR

```
### Instruction:
<Prompt>

### Input:
<Insert Context Here>

### Response:
```

## Contact
Kooten on discord