File size: 3,195 Bytes
a00264f |
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 |
---
license: other
license_name: deepseek
license_link: LICENSE
quantized_by: bartowski
pipeline_tag: text-generation
---
## Exllama v2 Quantizations of deepseek-coder-7b-instruct-v1.5
Using <a href="https://github.com/turboderp/exllamav2/releases/tag/v0.0.12">turboderp's ExLlamaV2 v0.0.12</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.
Original model: https://huggingface.co/deepseek-ai/deepseek-coder-7b-instruct-v1.5
No GQA - VRAM requirements will be higher
| Branch | Bits | lm_head bits | Size (4k) | Size (16k) | Description |
| -------------------------------------------------------------- | ---- | ------------ | --------- | ---------- | ----------- |
| [8_0](https://huggingface.co/Bartowski/deepseek-coder-7b-instruct-v1.5-exl2/tree/8_0) | 8.0 | 8.0 | 9.4 GB | 15.6 GB | Maximum quality that ExLlamaV2 can produce, near unquantized performance. |
| [6_5](https://huggingface.co/Bartowski/deepseek-coder-7b-instruct-v1.5-exl2/tree/6_5) | 6.5 | 8.0 | 8.6 GB | 14.8 GB | Near unquantized performance at vastly reduced size, **recommended**. |
| [5_0](https://huggingface.co/Bartowski/deepseek-coder-7b-instruct-v1.5-exl2/tree/5_0) | 5.0 | 6.0 | 7.2 GB | 13.4 GB | Slightly lower quality vs 6.5, but usable on 8GB cards with 4k context. |
| [4_25](https://huggingface.co/Bartowski/deepseek-coder-7b-instruct-v1.5-exl2/tree/4_25) | 4.25 | 6.0 | 6.5 GB | 12.7 GB | GPTQ equivalent bits per weight. |
| [3_5](https://huggingface.co/Bartowski/deepseek-coder-7b-instruct-v1.5-exl2/tree/3_5) | 3.5 | 6.0 | 5.9 GB | 12.1 GB | Lower quality, not recommended. |
## Download instructions
With git:
```shell
git clone --single-branch --branch 6_5 https://huggingface.co/bartowski/deepseek-coder-7b-instruct-v1.5-exl2 deepseek-coder-7b-instruct-v1.5-exl2-6_5
```
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 `deepseek-coder-7b-instruct-v1.5-exl2`:
```shell
mkdir deepseek-coder-7b-instruct-v1.5-exl2
huggingface-cli download bartowski/deepseek-coder-7b-instruct-v1.5-exl2 --local-dir deepseek-coder-7b-instruct-v1.5-exl2 --local-dir-use-symlinks False
```
To download from a different branch, add the `--revision` parameter:
Linux:
```shell
mkdir deepseek-coder-7b-instruct-v1.5-exl2-6_5
huggingface-cli download bartowski/deepseek-coder-7b-instruct-v1.5-exl2 --revision 6_5 --local-dir deepseek-coder-7b-instruct-v1.5-exl2-6_5 --local-dir-use-symlinks False
```
Windows (which apparently doesn't like _ in folders sometimes?):
```shell
mkdir deepseek-coder-7b-instruct-v1.5-exl2-6.5
huggingface-cli download bartowski/deepseek-coder-7b-instruct-v1.5-exl2 --revision 6_5 --local-dir deepseek-coder-7b-instruct-v1.5-exl2-6.5 --local-dir-use-symlinks False
```
Want to support my work? Visit my ko-fi page here: https://ko-fi.com/bartowski |