Loading the Mdeberta-v3-base

While loading the “microsoft/deberta-v3-base” model, I am getting unexpected warning message.

Code : model = AutoModelForTokenClassification.from_pretrained(“microsoft/deberta-v3-base”)

WARN : “Some weights of DebertaV2ForTokenClassification were not initialized from the model checkpoint at microsoft/deberta-v3-base and are newly initialized: [‘classifier.bias’, ‘classifier.weight’] You should probably TRAIN this model on a down-stream task to be able to use it for predictions and inference.”

???: Why AM I seeing "DebertaV2ForTokenClassification " while loading V3 model.

Is the DEVs called the print warn function with different args?:face_with_monocle::face_with_monocle::face_with_monocle:

1 Like

There are a number of possible causes, but the model hasn’t been updated for about two years, so if there is a problem, it’s probably because it hasn’t been updated since then, or because of changes to the library since then. Well, I think you can ignore the warnings and it will still work…:sweat_smile:

I am not sure you can see the jpeg i have attached.

the model config itself is pointing to V2.

I wondering how those 158 finetunes are using, is it V2 or V3?

if V3, it’s totally fine but if V2 those guys are doomed for real after realizing.

1 Like

Is there a way to report this to devs?

1 Like

Generally, if the person you are talking to is an individual, sending a mention with @+username is a good way to get their attention, but if the person you are talking to is an organization, mentions don’t work.
Also, since the person you are talking to is Microsoft and not Hugging Face, there is no way around it…

So, the most appropriate way to report this is to start a new discussion in the Discussion section.

However, there are cases where the code for the same model class can be reused, as is the case with Qwen 2 and 2.5. But V2 to V3 is probably a major version upgrade, so it’s probably still a bit strange.