bartowski/bagel-20b-v04-llama-exl2

Exllama v2 Quantizations of bagel-20b-v04-llama

Using turboderp's ExLlamaV2 v0.0.13 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./jondurbin/bagel-20b-v04-llama

Branch	Bits	lm_head bits	VRAM (4k)	VRAM (16k)	VRAM (32k)	Description
6_5	6.5	8.0	19.6 GB	21.0 GB	23.0 GB	Near unquantized performance at vastly reduced size, recommended.
4_25	4.25	6.0	13.8 GB	15.2 GB	17.2 GB	GPTQ equivalent bits per weight, slightly higher quality.
3_5	3.5	6.0	12.4 GB	13.8 GB	15.8 GB	Lower quality, only use if you have to.
3_0	3.0	6.0	11.1 GB	12.5 GB	15.5 GB	Very low quality. Usable on 12GB.

Download instructions

With git:

git clone --single-branch --branch 6_5 https://huggingface.co./bartowski/bagel-20b-v04-llama-exl2 bagel-20b-v04-llama-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 bagel-20b-v04-llama-exl2:

mkdir bagel-20b-v04-llama-exl2
huggingface-cli download bartowski/bagel-20b-v04-llama-exl2 --local-dir bagel-20b-v04-llama-exl2 --local-dir-use-symlinks False

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

Linux:

mkdir bagel-20b-v04-llama-exl2-6_5
huggingface-cli download bartowski/bagel-20b-v04-llama-exl2 --revision 6_5 --local-dir bagel-20b-v04-llama-exl2-6_5 --local-dir-use-symlinks False

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

mkdir bagel-20b-v04-llama-exl2-6.5
huggingface-cli download bartowski/bagel-20b-v04-llama-exl2 --revision 6_5 --local-dir bagel-20b-v04-llama-exl2-6.5 --local-dir-use-symlinks False

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