Stuck on an issue?

Lightrun Answers was designed to reduce the constant googling that comes with debugging 3rd party libraries. It collects links to all the places you might be looking at while hunting down a tough bug.

And, if you’re still stuck at the end, we’re happy to hop on a call to see how we can help out.

[examples] UserWarning: `max_length` is deprecated

See original GitHub issue

Not sure how many example scripts are affected by this:

src/transformers/generation_utils.py:963: UserWarning: `max_length` is deprecated in this function, use 
`stopping_criteria=StoppingCriteriaList(MaxLengthCriteria(max_length=max_length))` instead.

getting this with at least examples/pytorch/translation/run_translation.py

Issue Analytics

State:
Created 2 years ago
Comments:7 (7 by maintainers)

Top GitHub Comments

1reaction

Narsilcommented, May 3, 2021

Hi @stas00 ,

Yes, it’s my fault 1 deprecation too much on the generate function ! Just submitted a new PR to remove this extra warnings (which is incorrect).

max_length is used by generate but not by the subsequent functions.

1reaction

stas00commented, May 3, 2021

Thank you for this investigation, @bhadreshpsavani - that’s very helpful. It looks like the change was introduced just a day before I filed this Issue.

@Narsil, could we please check with you on this deprecation you introduced in https://github.com/huggingface/transformers/commit/aad95c7cdebc24e780e5a5cf39d832c015e40075

Unless I’m missing something the deprecation doesn’t seem to be complete since max_length is actively used in the same function:

https://github.com/huggingface/transformers/blob/4e7bf94e7280d2b725ac4644dbe9808560afa5d8/src/transformers/generation_utils.py#L919

I am not sure how a deprecated variable is still used normally in the logic…

Also it’s not documented as deprecated:

            max_length (:obj:`int`, `optional`, defaults to 20):
                The maximum length of the sequence to be generated.

it now generates warnings in the example scripts like run_translation.py and run_summarization.py - so if there is a new way could we please atomically adjust all the places that are now impacted by this change? The examples ideally should be in sync with API changes, since their purpose is to correctly demonstrate how to use the library.

The main entry point leading to this warning in seq2seq examples is:

https://github.com/huggingface/transformers/blob/4e7bf94e7280d2b725ac4644dbe9808560afa5d8/src/transformers/trainer_seq2seq.py#L161-L171

Thank you!

Top Results From Across the Web

max_length and stopping_criteria in generate() #17718 - GitHub

Accordingly, when you pass max_length you get the deprecated warning, which is ok - however, it recommends using the StoppingCriteriaList ...

How does max_length, padding and truncation arguments ...

padding=max_length will add a padding of 1 to the third sentence; truncate=True will truncate the first and second sentence so that their length ......

2101 (maxlength should be max_length) - Django's bug tracker

My personal recommendation would be to add the "max_length" parameter without removing "maxlength." This way, backwards-compatibility can be maintained while ...

Configuration options for isort

The max length of an import line (used for wrapping long imports). Type: Int ... 6-deprecated-alias-for-5, 7-noqa, 8-vertical-hanging-indent-bracket, ...

Trainer - Hugging Face

Trainer. The Trainer class provides an API for feature-complete training in PyTorch for most standard use cases. It's used in most of the...