Exllama v2 Quantizations of Mistral-7B-claude-chat

Using <a href="https://github.com/turboderp/exllamav2/releases/tag/v0.0.6">turboderp's ExLlamaV2 v0.0.6</a> for quantization.

Each branches contains an individual bits per weight.

Conversion was done using wikitext.parquet as calibration dataset.

Original model: https://huggingface.co/Norquinal/Mistral-7B-claude-chat

<a href="https://huggingface.co/bartowski/Mistral-7B-claude-chat-exl2/tree/4.0">4.0 bits per weight</a>

<a href="https://huggingface.co/bartowski/Mistral-7B-claude-chat-exl2/tree/8.0">8.0 bits per weight</a>

<a href="https://huggingface.co/bartowski/Mistral-7B-claude-chat-exl2/tree/6.0">6.0 bits per weight</a>

To download, you can use the following command:

With git:

git clone --single-branch --branch 4.0 https://huggingface.co/bartowski/Mistral-7B-claude-chat-exl2

With huggingface hub (credit to TheBloke for instructions):

pip3 install huggingface-hub

To download the main (only useful if you only care about measurement.json) branch to a folder called Mistral-7B-claude-chat-exl2:

mkdir Mistral-7B-claude-chat-exl2
huggingface-cli download bartowski/Mistral-7B-claude-chat-exl2 --local-dir Mistral-7B-claude-chat-exl2 --local-dir-use-symlinks False

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

mkdir Mistral-7B-claude-chat-exl2
huggingface-cli download bartowski/Mistral-7B-claude-chat-exl2 --revision 4.0 --local-dir Mistral-7B-claude-chat-exl2 --local-dir-use-symlinks False