Current CUDA Device does not support bfloat16. Please switch dtype to float16.
#4
by
marcos777123
- opened
Olá,
cuda version é 12.1 e cudnn 8.9.7
coloquei: model_dtype, torch_dtype = 'bf16', torch.float16 #torch.bfloat16
e está indo...