metadata
license: llama2
language:
- en
pipeline_tag: conversational
Another EXL2 version of AlpinDale's https://huggingface.co/alpindale/goliath-120b this one being at 2.37BPW.
Pippa llama2 Chat was used as the calibration dataset.
Can be run on two RTX 3090s w/ 24GB vram each.
Assuming Windows overhead, the following figures should be more or less close enough for estimation of your own use.
2.37BPW @ 4096 ctx
empty ctx
GPU split: 16/24
GPU1: 17.4/24GB
GPU2: 19.5/24GB
11~ tk/s
3000+ ctx
8~-12 tk/s