Remove Module._backend as it's not used anymore.

[gchanan]

Stack from ghstack:

• Kill non-shared cwrap tools. #25358 Kill non-shared cwrap tools.
• Update derivatives.yaml docs to refer to Declarations.yaml rather than Declarations.cwrap. #25357 Update derivatives.yaml docs to refer to Declarations.yaml rather than Declarations.cwrap.
• Get rid of extract_cwrap #25356 Get rid of extract_cwarp.
• Get rid of more unused plugins. #25355 Get rid of more unused plugins.
• Get rid of torch._thnn. #25354 Get rid of torch._thnn.
• Stop doing nn wrap. #25353 Stop doing nn wrap.
• Stop initializing THNN backend. #25352 Stop initializing THNN backend.
• [JIT] Attempt to enable CrossMapLRN2d, as it no longer uses Module._backend. #25343 [JIT] Attempt to enable CrossMapLRN2d, as it no longer uses Module._backend.
• Remove Module._backend as it's not used anymore. #25342 Remove Module._backend as it's not used anymore.
• Move autograd function for CrossMapLRN2d from being backend specific to modules/_functions. #25339 Move autograd function for CrossMapLRN2d from being backend specific to modules/_functions.
• Kill backend-specific lookup in CrossMapLRN2d, as it never succeeds. #25331 Kill backend-specific lookup in CrossMapLRN2d, as it never succeeds.
• Kill ConvTransposeMixin.forward, which doesn't seem to be used. #25326 Kill ConvTransposeMixin.forward, which doesn't seem to be used.
• Remove THNN sparse autograd Functions. #25323 Remove THNN sparse autograd Functions.
• Kill THNN function auto generation. #25322 Kill THNN function auto generation.

Differential Revision: D17101571

[facebook-github-bot]

@gchanan merged this pull request in 23fde77.

[toosi]

Hi @gchanan,

In my code, I use _backend.SpatialFullConvolution_updateOutput and now that in the latest version of Pytorch THNN is not supported anymore, I wonder if you could point me to a substitute if there exist any?
(I need access to the feedback parameters of the convolution because I am working on alternative learning algorithms to Backprop.)
my question in forum

Thanks,
Tahereh

[gchanan]

SpatialFullConvolution_updateOutput doesn't exist anymore, so I don't think that code would even get executed. What were you trying to do?