Make it a 22b
#1
by
jackboot
- opened
Since it's a merge, add some attention heads and make a bigger model: https://huggingface.co/chargoddard/llama2-22b
Since it's a merge, add some attention heads and make a bigger model: https://huggingface.co/chargoddard/llama2-22b
Do you expect any effects from it beside the higher demands on hardware?
Better performance. Since it's not done perhaps testing both would settle it.