Edit model card

Exllama v2 Quantizations of gemma-2-9b-it

Using turboderp's ExLlamaV2 v0.1.7 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/google/gemma-2-9b-it

Prompt format

<bos><start_of_turn>user
{prompt}<end_of_turn>
<start_of_turn>model
<end_of_turn>
<start_of_turn>model

Note that this model does not support a System prompt.

Available sizes

Branch Bits lm_head bits VRAM (4k) VRAM (16k) VRAM (32k) Description
8_0 8.0 8.0 11.9 GB 15.9 GB 21.3 GB Maximum quality that ExLlamaV2 can produce, near unquantized performance.
6_5 6.5 8.0 10.4 GB 14.4 GB 19.8 GB Very similar to 8.0, good tradeoff of size vs performance, recommended.
5_0 5.0 6.0 8.6 GB 12.6 GB 18.0 GB Slightly lower quality vs 6.5, but usable on 8GB cards.
4_25 4.25 6.0 7.9 GB 11.9 GB 17.3 GB GPTQ equivalent bits per weight, slightly higher quality.
3_5 3.5 6.0 7.1 GB 11.1 GB 16.9 GB Lower quality, only use if you have to.

Download instructions

With git:

git clone --single-branch --branch 6_5 https://huggingface.co/bartowski/gemma-2-9b-it-exl2 gemma-2-9b-it-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/gemma-2-9b-it-exl2 --revision 6_5 --local-dir gemma-2-9b-it-exl2-6_5

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

huggingface-cli download bartowski/gemma-2-9b-it-exl2 --revision 6_5 --local-dir gemma-2-9b-it-exl2-6.5

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

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Examples
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated) instead.