FutureWarning: `torch.cuda.amp.GradScaler(args...)` is deprecated.

[si-open]

train.py:196: FutureWarning: torch.cuda.amp.GradScaler(args...) is deprecated. Please use torch.amp.GradScaler('cuda', args...) instead.
scaler = torch.cuda.amp.GradScaler(enabled=(dtype == 'float16'))
num decayed parameter tensors: 26, with 10,755,456 parameters
num non-decayed parameter tensors: 13, with 4,992 parameters
using fused AdamW: True
compiling the model... (takes a ~minute)
W0903 20:13:19.043000 55763 torch/_inductor/utils.py:1436] [0/0] Not enough SMs to use max_autotune_gemm mode