Inference or finetuning with RTX 3000 / A100 on CUDA 11

[tianjianh]

Hey I am just wondering is there way to load any pre-trained models with A100 on CUDA 11? It seems that deepspeed==0.4.4 and triton==0.4.2 do not work with CUDA 11 but pre-trained models require those old versions of deepspeed and triton. Thanks in advance!