TPU train;how to fix this bug;runtime error: Numpy is not available

xla
1 
import numpy as np
import torch
print(torch.__version__)
print(np.__version__)
a = np.array([1, 2, 3])
t = torch.from_numpy(a)

1.14.0a0+git8cb5c55
1.19.5

RuntimeError Traceback (most recent call last)
/tmp/ipykernel_2383/3624488259.py in
4 print(np.version)
5 a = np.array([1, 2, 3])
----> 6 t = torch.from_numpy(a)
7

RuntimeError: Numpy is not available
I tried several version :1.12,1.13 ,but it didn’t work

2

Thanks for reporting! Do you mind open a github issue under Issues · pytorch/xla · GitHub? We now handles all TPU and pytorch/xla issue there.