bartowski's picture
Update README.md
d6e74e0 verified
metadata
pipeline_tag: text-generation
license: other
quantized_by: bartowski

Update Jan 27: This has been redone with the proper token mappings and rope scaling, performance seems improved, please comment if not

Exllama v2 Quantizations of internlm2-chat-20b-llama-test

Using turboderp's ExLlamaV2 v0.0.12 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/internlm/internlm2-chat-20b

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/internlm2-chat-20b-llama-exl2 internlm2-chat-20b-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 internlm2-chat-20b-llama-exl2:

mkdir internlm2-chat-20b-llama-exl2
huggingface-cli download bartowski/internlm2-chat-20b-llama-exl2 --local-dir internlm2-chat-20b-llama-exl2 --local-dir-use-symlinks False

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

Linux:

mkdir internlm2-chat-20b-llama-exl2-6_5
huggingface-cli download bartowski/internlm2-chat-20b-llama-exl2 --revision 6_5 --local-dir internlm2-chat-20b-llama-exl2-6_5 --local-dir-use-symlinks False

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

mkdir internlm2-chat-20b-llama-exl2-6.5
huggingface-cli download bartowski/internlm2-chat-20b-llama-exl2 --revision 6_5 --local-dir internlm2-chat-20b-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