centerformer icon indicating copy to clipboard operation
centerformer copied to clipboard
verified publisher icon TuSimple

AUTOMATIC MIXED PRECISION

Open zhaowenZhou opened this issue 2 years ago • 0 comments

Has anyone tried torch.cuda.amp? Seems that ms_attention doesn't support fp16 even after I modified ms_deform_attn_forward_cuda Any other way to implement amp? Or is there any ways to reduce the GPU memory? I got cuda OOM for bs=4 every time

zhaowenZhou avatar May 25 '23 10:05 zhaowenZhou