mirror of https://github.com/coqui-ai/TTS.git
numpy to use CPU when using CUDA
parent
95385c8797
commit
61a1d59ac5
|
@ -157,7 +157,7 @@ def check_update(model, grad_clip, ignore_stopnet=False):
|
|||
grad_norm = torch.nn.utils.clip_grad_norm_([param for name, param in model.named_parameters() if 'stopnet' not in name], grad_clip)
|
||||
else:
|
||||
grad_norm = torch.nn.utils.clip_grad_norm_(model.parameters(), grad_clip)
|
||||
if np.isinf(grad_norm):
|
||||
if np.isinf(grad_norm.cpu()):
|
||||
print(" | > Gradient is INF !!")
|
||||
skip_flag = True
|
||||
return grad_norm, skip_flag
|
||||
|
|
Loading…
Reference in New Issue