Allow forward pass without gradient

chr5tphr / zennit

Zennit is a high-level framework in Python using PyTorch for explaining/exploring neural networks using attribution methods like LRP.

Other

191 stars 32 forks source link

Closed chr5tphr closed 3 years ago

chr5tphr commented 3 years ago

previously, a crash would occur if a module with a registered rule would be passed without requiring gradient
this commit explicitly allows forward of a module with a registered rule without requiring gradient computation