[docs] optimizers, hyperparam search, training features#44290
[docs] optimizers, hyperparam search, training features#44290stevhliu merged 7 commits intohuggingface:mainfrom
Conversation
|
The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update. |
SunMarc
left a comment
SunMarc
left a comment
There was a problem hiding this comment.
Thanks for these, this is already much better, left a couple of comments
| @@ -0,0 +1,255 @@ | |||
| <!--Copyright 2026 The HuggingFace Team. All rights reserved. | |||
There was a problem hiding this comment.
we can keep these here for now, but i feel like they can be moved in other sections. Trainer receipe name don't fully represent what we have here. Maybe we can think of another name ? I feel like we could create another kind of recipes here.
There was a problem hiding this comment.
hmm yeah a recipe is more of a contained end-to-end pattern. i'll rename this to "Trainer features" which i think is more accurate of what is described here. this would free up Trainer recipes for actual recipes later
stevhliu
force-pushed
the
training-part-3
branch
from
9c3c28a to
f8d060b
Compare
|
i updated the scheduler docs to include |
stevhliu
force-pushed
the
training-part-3
branch
from
f8d060b to
c22682f
Compare
|
updated with flashoptim from #44438! |
stevhliu
force-pushed
the
training-part-3
branch
from
c22682f to
60471db
Compare
…44290) * optimizer + scheduler * hyperparam * training features * feedback * fix * greedylr * flashoptim
3 participants
trainer_recipes.mdto show how to use other practicalTrainerfeatures outside of the basic training loopoptimizers.mdwith how to customize it (prebuilt instances, passing a class + kwargs, subclassing)