PAC_CRF step setting question
GITSHOHOKU opened this issue · comments
Hi
Should we set CRF step to 1 when training the whole model? And step to whatever we want when in the inference processing?
Because when I set CRF step to num>1, it will cause the in-place operation in the training process.
We train with the same number of iters used in inference (e.g. 5). What's the in-place operation you referred to?
We train with the same number of iters used in inference (e.g. 5). What's the in-place operation you referred to?
The inplace_operation mentioned before points to the iteration manipulate value logQ , but it never happened again. Just a warning:
RuntimeWarning: Iterating over a tensor might cause the trace to be incorrect. Passing a tensor of different shape won't change the number of iterations executed (and might lead to errors or silently give incorrect results).
which doesn't affect the algorithm running.
Maybe it was just caused by Pytorch's inner mechanism. Now the training step is well.
Thanks for your quick reply.