How to debug "gradient computation has been modified by an inplace operation" errors?

[KFrank]

Hi Ume!

@ume_habiba

one of the variables needed for gradient computation has been modified by an inplace operation:!

In addition to Alex’s above post – in particular his suggestion to use
set_detect_anomaly() – here is a post that illustrates some ways
to debug these kinds of errors:

"RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation: [torch.FloatTensor [64, 1]], which is output 0 of AsStridedBackward0, is at version 3; expected version 2 instead. Hint: the backtrace further a

Hi Fahmyadan and Sangyoon! Here are some suggestions about how to track down (and maybe fix) inplace-modification errors. Note that an inplace modification in the forward pass is not necessarily* an error – it depends on whether and how the tensor that was modified is used in the backward pass. Note that inplace operations can be useful for saving memory – if you replace an innocent inplace operation with an out-of-place equivalent, your training will use more memory (and, to a minor e…

Good luck!

K. Frank