File size: 3,282 Bytes
6f23c72
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
33ccdcb
 
 
 
 
 
 
6f23c72
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
---

license: other
license_name: deepseek
license_link: https://github.com/deepseek-ai/DeepSeek-Coder/blob/main/LICENSE-MODEL
tags:
- merge
quantized_by: bartowski
pipeline_tag: text-generation
---

## Exllama v2 Quantizations of Everyone-Coder-33b-v2-Base

Using <a href="https://github.com/turboderp/exllamav2/releases/tag/v0.0.13">turboderp's ExLlamaV2 v0.0.13</a> for quantization.

## The "main" branch only contains the measurement.json, download one of the other branches for the model (see below)

Each branch contains an individual bits per weight, with the main one containing only the meaurement.json for further conversions.

Conversion was done using the default calibration dataset.

Default arguments used except when the bits per weight is above 6.0, at that point the lm_head layer is quantized at 8 bits per weight instead of the default 6.

Original model: https://huggingface.co/rombodawg/Everyone-Coder-33b-v2-Base

| Branch | Bits | lm_head bits | VRAM (4k) | VRAM (16k) | VRAM (32k) | Description |
| ------ | ---- | ------------ | ---- | ---- | ---- | ----------- |
| [6_5](https://huggingface.co/bartowski/Everyone-Coder-33b-v2-Base-exl2/tree/6_5)   | 6.5  | 8.0 | 28.9 GB | 31.6 GB | 35.6 GB | Near unquantized performance at vastly reduced size, **recommended**.       |
| [4_25](https://huggingface.co/bartowski/Everyone-Coder-33b-v2-Base-exl2/tree/4_25) | 4.25 | 6.0 | 19.5 GB | 22.2 GB | 26.2 GB | GPTQ equivalent bits per weight, slightly higher quality.                   |
| [3_5](https://huggingface.co/bartowski/Everyone-Coder-33b-v2-Base-exl2/tree/3_5)   | 3.5  | 6.0 | 16.5 GB | 19.2 GB | 23.2 GB | Lower quality, only use if you have to.                                     |
| [3_0](https://huggingface.co/bartowski/Everyone-Coder-33b-v2-Base-exl2/tree/3_0)   | 3.0  | 6.0 | 14.3 GB | 17.0 GB | 21.0 GB | Very low quality, usable with 16gb of VRAM. |
| [2_4](https://huggingface.co/bartowski/Everyone-Coder-33b-v2-Base-exl2/tree/3_0)   | 2.4  | 6.0 | 12.0 GB | 14.7 GB | 18.7 GB | Extremely low quality, only recommended if trying to fit into 12GB with under 4k context. |

## Download instructions

With git:

```shell
git clone --single-branch --branch 6_5 https://huggingface.co/bartowski/Everyone-Coder-33b-v2-Base-exl2
```

With huggingface hub (credit to TheBloke for instructions):

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

To download the `main` (only useful if you only care about measurement.json) branch to a folder called `Everyone-Coder-33b-v2-Base-exl2`:

```shell
mkdir Everyone-Coder-33b-v2-Base-exl2
huggingface-cli download bartowski/Everyone-Coder-33b-v2-Base-exl2 --local-dir Everyone-Coder-33b-v2-Base-exl2 --local-dir-use-symlinks False
```

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

Linux:

```shell
mkdir Everyone-Coder-33b-v2-Base-exl2-6_5
huggingface-cli download bartowski/Everyone-Coder-33b-v2-Base-exl2 --revision 6_5 --local-dir Everyone-Coder-33b-v2-Base-exl2-6_5 --local-dir-use-symlinks False
```

Windows (which apparently doesn't like _ in folders sometimes?):

```shell
mkdir Everyone-Coder-33b-v2-Base-exl2-6.5
huggingface-cli download bartowski/Everyone-Coder-33b-v2-Base-exl2 --revision 6_5 --local-dir Everyone-Coder-33b-v2-Base-exl2-6.5 --local-dir-use-symlinks False
```