Exllama v2 Quantizations of Qwen1.5-14B-Chat
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/Qwen/Qwen1.5-14B-Chat/
No GQA - VRAM requirements will be higher
| Branch | Bits | lm_head bits | VRAM (4k) | VRAM (16k) | Description |
|---|---|---|---|---|---|
| 8_0 | 8.0 | 8.0 | 18.5 GB | 28.8 GB | Near unquantized performance at vastly reduced size, recommended. |
| 6_5 | 6.5 | 8.0 | 15.9 GB | 26.4 GB | Near unquantized performance at vastly reduced size, recommended. |
| 5_0 | 5.0 | 6.0 | 13.3 GB | 23.9 GB | Slightly lower perplexity vs 6.5. |
| 4_25 | 4.25 | 6.0 | 12.0 GB | 22.6 GB | GPTQ equivalent bits per weight, can fit in 12 GB card with lower context. |
| 3_75 | 3.75 | 6.0 | 11.1 GB | 21.7 GB | Lower quality but still generally usable. |
| 3_0 | 3.0 | 6.0 | 10.0 GB | 20.6 GB | Very low quality, not recommended unless you have to. |
VRAM requirements listed for both 4k context and 16k context since without GQA the differences are massive (9.6 GB)
Download instructions
With git:
git clone --single-branch --branch 6_5 https://huggingface.co/bartowski/Qwen1.5-14B-Chat-exl2 Qwen1.5-14B-Chat-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 Qwen1.5-14B-Chat-exl2:
mkdir Qwen1.5-14B-Chat-exl2
huggingface-cli download bartowski/Qwen1.5-14B-Chat-exl2 --local-dir Qwen1.5-14B-Chat-exl2 --local-dir-use-symlinks False
To download from a different branch, add the --revision parameter:
Linux:
mkdir Qwen1.5-14B-Chat-exl2-6_5
huggingface-cli download bartowski/Qwen1.5-14B-Chat-exl2 --revision 6_5 --local-dir Qwen1.5-14B-Chat-exl2-6_5 --local-dir-use-symlinks False
Windows (which apparently doesn't like _ in folders sometimes?):
mkdir Qwen1.5-14B-Chat-exl2-6.5
huggingface-cli download bartowski/Qwen1.5-14B-Chat-exl2 --revision 6_5 --local-dir Qwen1.5-14B-Chat-exl2-6.5 --local-dir-use-symlinks False
Want to support my work? Visit my ko-fi page here: https://ko-fi.com/bartowski