[–] brucethemoose2 3y ago ↗ A 48GB Nvidia GPU for LLaMa 65B. Maybe 2x24GB?Less for LLaMA 33B, which is still very good.Use a repo that supports QLORA, like https://github.com/OpenAccess-AI-Collective/axolotl
1 comment
[ 3.6 ms ] story [ 20.2 ms ] threadLess for LLaMA 33B, which is still very good.
Use a repo that supports QLORA, like https://github.com/OpenAccess-AI-Collective/axolotl