File size: 935 Bytes
60aacdc
 
 
 
 
 
 
 
 
6696709
60aacdc
 
 
 
 
 
 
 
 
 
 
 
6696709
60aacdc
 
 
 
 
 
 
 
 
 
 
 
6696709
60aacdc
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
---
license: apache-2.0
tags:
- finetuned
- Instruct
- code
quantized_by: bartowski
---

# Exllama v2 Quantizations of KAI-7B-Instruct at 8.0,6.0,4.0

Using <a href="https://github.com/turboderp/exllamav2/releases/tag/v0.0.7">turboderp's ExLlamaV2 v0.0.7</a> for quantization.

Conversion was done using wikitext-103-raw-v1-test.parquet as calibration dataset.

Original model: https://huggingface.co/Keynote-Technology/KAI-7B-Instruct

## Download instructions

With git:

```shell
git clone --single-branch --branch 8.0,6.0,4.0 https://huggingface.co/bartowski/KAI-7B-Instruct-exl2
```

With huggingface hub (credit to TheBloke for instructions):

```shell
pip3 install huggingface-hub
```

To download from a different branch, add the `--revision` parameter:

```shell
mkdir KAI-7B-Instruct-exl2
huggingface-cli download bartowski/KAI-7B-Instruct-exl2 --revision  --local-dir KAI-7B-Instruct-exl2 --local-dir-use-symlinks False
```