cc_staff
46
edits
PyTorch: Difference between revisions
Jump to navigation
Jump to search
no edit summary
No edit summary |
No edit summary |
||
| Line 109: | Line 109: | ||
To enable or disable TF32 on <code>torch >= 1.12.0</code> set the following flags to <code>True</code> or <code>False</code> accordingly: | To enable or disable TF32 on <code>torch >= 1.12.0</code> set the following flags to <code>True</code> or <code>False</code> accordingly: | ||
<!--T:551--> | |||
torch.backends.cuda.matmul.allow_tf32 = False # Enable/disable TF32 for matrix multiplications | torch.backends.cuda.matmul.allow_tf32 = False # Enable/disable TF32 for matrix multiplications | ||
torch.backends.cudnn.allow_tf32 = False # Enable/disable TF32 for convolutions | torch.backends.cudnn.allow_tf32 = False # Enable/disable TF32 for convolutions | ||