Branch	Bits	lm_head bits	Size	Description
8_0	8.0	8.0	9.8 GB	Maximum quality that ExLlamaV2 can produce, near unquantized performance.
6_5	6.5	8.0	8.6 GB	Very similar to 8.0, good tradeoff of size vs performance, recommended.
5_0	5.0	6.0	7.4 GB	Slightly lower quality vs 6.5, but usable on 8GB cards.
4_25	4.25	6.0	6.7 GB	GPTQ equivalent bits per weight, slightly higher quality.
3_5	3.5	6.0	6.1 GB	Lower quality, only use if you have to.

All VRAM requirements estimated from 16k context. For 32k context add ~2 GB.

Download instructions

With git:

git clone --single-branch --branch 6_5 https://huggingface.co/bartowski/zephyr-7b-dpo-full-exl2 zephyr-7b-dpo-full-exl2-6_5

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 zephyr-7b-dpo-full-exl2:

shell

mkdir zephyr-7b-dpo-full-exl2
huggingface-cli download bartowski/zephyr-7b-dpo-full-exl2 --local-dir zephyr-7b-dpo-full-exl2 --local-dir-use-symlinks False

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

Linux:

shell

mkdir zephyr-7b-dpo-full-exl2-6_5
huggingface-cli download bartowski/zephyr-7b-dpo-full-exl2 --revision 6_5 --local-dir zephyr-7b-dpo-full-exl2-6_5 --local-dir-use-symlinks False

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

shell

mkdir zephyr-7b-dpo-full-exl2-6.5
huggingface-cli download bartowski/zephyr-7b-dpo-full-exl2 --revision 6_5 --local-dir zephyr-7b-dpo-full-exl2-6.5 --local-dir-use-symlinks False