5.0.0 `tie_embeddings_and_encoder_decoder` removed without indication

[AmitMY]

System Info

• transformers version: 5.0.0.dev0 (main branch)
• Platform: Linux
• Python version: 3.12
• PyTorch version: 2.x with CUDA
• GPU: NVIDIA (tested)

Who can help?

@stevhliu

Information

• The official example scripts
• My own modified scripts

Tasks

• An officially supported task in the examples folder (such as GLUE/SQuAD, ...)
• My own task or dataset (give details below)

Reproduction

In version 4 of transformers, there's a method tie_embeddings_and_encoder_decoder.
This method is not present in version 5, without a deprecation note in the release notes.

Run:

from transformers import AutoTokenizer, AutoModelForMaskedLM

model_id = "answerdotai/ModernBERT-base"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForMaskedLM.from_pretrained(model_id)

model.tie_embeddings_and_encoder_decoder()

Expected behavior

deprecation notice / docs what to do instead

[Tanuj-Taneja1]

Hi @AmitMY , You are right about the issue I was able to reproduce it in the main branch. I couldnt find the 5.0.0.dev0 version though.

I think it was removed in commit 6f6095e
In PR : #41580.

Hi @ArthurZucker, could you provide any insights on why the method tie_embeddings_and_encoder_decoder was removed so we can add a warning and depreciated message regarding it.

[Tanuj-Taneja1]

Also this seems like a good first issue so I would be happy to work on it. I have previously contributed to other open source projects but not hugging face yet so it could be a great start for me. Feel free to assign me the issue

[ArthurZucker]

Hey heyy! This is just because it's the same as just tie_embeddings! We can keep just a reference to it, its just that we merged tie_encoder_decoder and tie_words_embeddings. This was done in #42815

[AmitMY]

For me, a comment in the release notes would suffice as a solution.

Removed tie_embeddings_and_encoder_decoder in favor of tie_embeddings