Update autograd metadata tracking whether the given Tensor was modified in place.

This is to enable more accurate error checking within the autograd engine. It is already done automatically by PyTorch functions and within custom Function when mark_dirty() is called appropriately so you only need to call this explicitly if you are doing inplace operation on the Tensor data in a way that Pytorch doesn’t know about. For example a custom kernel that reads the Tensor data_ptr and modifies the memory inplace based on this pointer.

Note that incrementing the version counter multiple times for a single inplace operation is not problematic.


Access comprehensive developer documentation for PyTorch

View Docs


Get in-depth tutorials for beginners and advanced developers

View Tutorials


Find development resources and get your questions answered

View Resources