Repository navigation
support for aot (ahead of time) to deal with ml.net models #6942
Description
is there api to load ml.net model and predict with no dynamic code generation (reflection)?
Describe the solution you'd like
ml.net api that is compatible with aot principles.
Reacted by Dani Santos, Vicente Penades, AFCMS and Kelvin
Activity
[-]support from aot (ahead of time) api[/-][+]support for aot (ahead of time) to deal with mi.net models[/+][-]support for aot (ahead of time) to deal with mi.net models[/-][+]support for aot (ahead of time) to deal with ml.net models[/+]michaelgsharp commentedon Jan 10, 2024
Currently we are not AOT compatible, though this is something we have talked about looking into.
@luisquintanilla is this something we are thinking about doing still? Or is it further back in our backlog?
luisquintanilla commentedon Jan 10, 2024
Thanks for this suggestion @NofAssiri. Could you please provide additional details about your use case.
I think we still need to understand what the cost / benefits of this are and get more feedback around scenarios. At this time, there are no immediate plans for this work and is further back in our backlog.
FrancescoCarraro commentedon May 8, 2024
my 2 cents. It seems to me benefits of AOT are clear: computational time is very different and this is extremely important in some scenarios. I'm currently working on a couple of apps (a console app and a MAUI app) for being used in scientific analysis and having ML.NET compatible with AOT compilation would be really great to have shorter waitings for results. I really hope you would implement AOT support!
vpenades commentedon Feb 16, 2026
Just hit this issue, as of 2026 it seems ML is not AOT ready.
We need AOT for cross platform support.
This is not optional.
rosebyte commentedon Oct 22, 2026
Triage: there are no plans to support AOT right now, if we will have resources and funding to implement it in future, we will create a new issue for it.