File size: 1,014 Bytes
1720bbb 7549996 1720bbb 7549996 1720bbb 7549996 1720bbb |
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 |
---
license: other
license_name: deepseek
license_link: LICENSE
quantized_by: bartowski
---
# Exllama v2 Quantizations of deepseek-coder-6.7b-instruct at 8.0 bits per weight
Using <a href="https://github.com/turboderp/exllamav2/releases/tag/master">turboderp's ExLlamaV2 master</a> for quantization.
Conversion was done using Evol-Instruct-Code-80k-v1.parquet as calibration dataset.
Original model: https://huggingface.co/deepseek-ai/deepseek-coder-6.7b-instruct
## Download instructions
With git:
```shell
git clone --single-branch --branch 8.0 https://huggingface.co/bartowski/deepseek-coder-6.7b-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 deepseek-coder-6.7b-instruct-exl2
huggingface-cli download bartowski/deepseek-coder-6.7b-instruct-exl2 --revision 8_0 --local-dir deepseek-coder-6.7b-instruct-exl2 --local-dir-use-symlinks False
```
|