Remove unsupported and confusing "OpenLlama" architecture#24913
Remove unsupported and confusing "OpenLlama" architecture#24913tomaarsen wants to merge 1 commit intohuggingface:mainfrom
Conversation
This reverts commit c2c99dc.
|
The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. |
|
No we cannot removing the architecture entirely that would be a breaking change in the library that would be unprecedented and against our philosphy/commitment to backward compatibility. Even removing it entirely from the docs seem a bit extreme. If users are confused by which class to use to load a model, they should use the Auto model API to let it select the right class for them. We can add a disclaimer in the doc page of open-llama doc or move it to the deprecated models but that's pretty much the extent of what we can do. cc @LysandreJik |
|
I had such a fear - I understand your reasoning completely. Whichever way we go, we end up with a suboptimal situation. I'll let the user on Discord know that it won't be changed. I could add a disclaimer that the OpenLlama models do not use the OpenLlama architecture, but simply the Llama one. |
Yes, that would be great! |
This reverts commit c2c99dc from #22795.
Hello!
What does this PR do?
Removes Open-Llama, a confusingly named and unused model architecture from which all documentation links throw 404 errors.
Motivation
OpenLlamaForCausalLM.OpenLlamaForCausalLM.LlamaForCausalLMinstead.In case you are opposed out of principle, i.e. that no architectures should be removed, then we may want to at least update the documentation. However, I urge you to remove this confusing architecture.
Before submitting
Pull Request section?
to it if that's the case.
documentation guidelines, and
here are tips on formatting docstrings.
Who can review?
Reviewers of the original PR: @ArthurZucker @sgugger
cc: @s-JoL