Hi @xiuqhou! Congratulations on the paper, awesome work! And thanks for working on transformers implementation! Feel free to ping me when it's ready for review or if you have any questions!

Hi @qubvel Thanks for your support! The code is now ready for review—I'd greatly appreciate it if you could take a look and share your feedback. Please let me know if there’s anything that needs improvement.

@ArthurZucker @qubvel @stevhliu Sorry for the late reply! Thanks for your detailed review and suggestions. I have pushed commits according to the reviews except #34900 (comment) about the modular style.
I have finished converting the code to the modular style but met a bug #36208 in utils/modular_model_converter.py.
Once it is fixed, I will commit this part of code.

CI errors seem not related to changes.

Hey @xiuqhou thanks for working on the PR and opening the issue!

I think we can use the following workaround while the issue is not fixed.
Just override the function in modular file to avoid local imports (we can leave numpy and torch only)

def get_numpy_to_framework_fn(arr) -> Callable:
    """
    Returns a function that converts a numpy array to the framework of the input array.

    Args:
        arr (`np.ndarray`): The array to convert.
    """
    if isinstance(arr, np.ndarray):
        return np.array
    elif is_torch_available() and is_torch_tensor(arr):
        return torch.tensor
    else:
        raise RuntimeError("...")

Let me know if that works for you!

@qubvel thanks for your suggestion!

I tried to overwrite the function in modular_relation_detr.py but the error still exists. If I override the original function inherited from src/transformers/models/detr/image_processing_detr.py, it will work. But this will change the code of other models, which is beyond the code scope of this PR.

I have a question about whether this function is necessary. I searched all the code of transformers, and found that this function has been defined many times in DETR related models, but it has never been called. Can we remove this unused function?

Oh, it might be a redundant function that was copied from one model to another 😄 In that case we can open a separate PR to remove it everywhere

Hey @xiuqhou , thanks so much for your work so far! Since this will be (most likely) the new best DETR model on HF, I wanted to ask if you think that a merge is likely in the upcoming months? I saw that the issue you mentioned #36208 was fixed by #36279

Hi @daniel-bogdoll Sorry for the late reply! As #36208 has been fixed, I have committed code about the modular style for relation-detr. Now I think the PR is fully prepared for review or merge. Please let me know if there are any remaining improvements needed before merging.