bartowski/DiscoPOP-zephyr-7b-gemma-exl2

Exllama v2 Quantizations of DiscoPOP-zephyr-7b-gemma

Using turboderp's ExLlamaV2 v0.1.5 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./SakanaAI/DiscoPOP-zephyr-7b-gemma

Prompt format

<bos><|im_start|>system
{system_prompt}<|im_end|>
<|im_start|>user
{prompt}<|im_end|>
<|im_start|>assistant

Available sizes

No GQA - VRAM requirements will be higher

Branch	Bits	lm_head bits	Size (4k)	Size (16k)	Description
8_0	8.0	8.0	14.0 GB	19.4 GB	Maximum quality that ExLlamaV2 can produce, near unquantized performance.
6_5	6.5	8.0	12.5 GB	17.9 GB	Near unquantized performance at vastly reduced size, recommended.
5_0	5.0	6.0	10.9 GB	16.3 GB	Slightly lower quality vs 6.5, but usable on 8GB cards with 4k context.
4_25	4.25	6.0	10.2 GB	15.7 GB	GPTQ equivalent bits per weight.
3_5	3.5	6.0	9.5 GB	14.9 GB	Lower quality, not recommended.

Download instructions

With git:

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

With huggingface hub (credit to TheBloke for instructions):

pip3 install huggingface-hub

To download a specific branch, use the --revision parameter. For example, to download the 6.5 bpw branch:

Linux:

huggingface-cli download bartowski/DiscoPOP-zephyr-7b-gemma-exl2 --revision 6_5 --local-dir DiscoPOP-zephyr-7b-gemma-exl2-6_5

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

huggingface-cli download bartowski/DiscoPOP-zephyr-7b-gemma-exl2 --revision 6_5 --local-dir DiscoPOP-zephyr-7b-gemma-exl2-6.5

Want to support my work? Visit my ko-fi page here: https://ko-fi.com/bartowski