Thanks ! It’s quite strange to have an error when the package I am using is stable, older version. I have
been running it several weeks without problem.
But I see something is still not right. Today I tried on Kaggle :
!pip install fastai2
Successfully installed dataclasses-0.6 fastai2-0.0.30 fastcore-0.1.38 torch-1.6.0 torchvision-0.7.0
I can run pass the above errors. But another error appears when I use learn.to_fp16() :
/opt/conda/lib/python3.7/site-packages/fastai2/callback/fp16.py in before_fit(self)
84 def before_fit(self):
—> 85 assert self.dls.device.type == ‘cuda’, "Mixed-precision training requires a GPU, remove the call to_fp16"
86 if self.learn.opt is None: self.learn.create_opt()
87 self.model_pgs,self.master_pgs = get_master(self.opt, self.flat_master)
AssertionError: Mixed-precision training requires a GPU, remove the call to_fp16
This is surprising since I already turned on GPU mode on Kaggle