Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Easy path to using anything with a ChainRulesCore.RuleConfig #11

Closed
oxinabox opened this issue Oct 1, 2021 · 4 comments
Closed

Easy path to using anything with a ChainRulesCore.RuleConfig #11

oxinabox opened this issue Oct 1, 2021 · 4 comments
Labels
feature New feature or request

Comments

@oxinabox
Copy link
Member

oxinabox commented Oct 1, 2021

We have talked about this before, just writing it down so it doesn't get lost.
The ChainRulesCore.RuleConfig with the right traits, provides all the parts needed for either getting the pushforward (frule_via_ad) or pullback (rrule_via_ad); or both.
This primitives are enough to let AbstractDifferentiation.jl define everything else.

Any ChainRules supporting AD that supports rules that require calling back into AD must define one of these.
Zygote.ZygoteRuleConfig and Diffractor.DiffractorRuleConfig for example.

Given such a RuleConfig type, it should be a one-liner to get the whole AbstractDifferentiation API.
We should provide a macro to make that so.

@gdalle gdalle added the feature New feature or request label Oct 5, 2023
@gdalle
Copy link
Member

gdalle commented Oct 5, 2023

I think this is now ReverseRuleConfigBackend?

@schlichtanders
Copy link

Is there some progress here?

@gdalle
Copy link
Member

gdalle commented Feb 11, 2024

I think it should be ReverseRuleConfigBackend? Can you post a MWE if what you're trying does not work?

@oxinabox
Copy link
Member Author

Yeah this is basically done.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
feature New feature or request
Projects
None yet
Development

No branches or pull requests

3 participants