The documentation of transformers.generation_utils.GenerationMixin.generate doesn't match the code of it

[zhaowei-wang-nlp]

Environment info

• transformers version: 4.17.0
• Platform: Linux
• Python version: 3.8.12
• PyTorch version (GPU?): 1.7.1
• Tensorflow version (GPU?):
• Using GPU in script?:
• Using distributed or parallel set-up in script?:

Who can help

@sgugger@patrickvonplaten @Narsil

Information

The documentation of transformers.generation_utils.GenerationMixin.generate shows many parameters have default value, but the code of it shows all parameters are set to be None by default.

The problem arises when using:

• the official example scripts: (give details below)
• my own modified scripts: (give details below)

The tasks I am working on is:

• an official GLUE/SQUaD task: (give the name)
• my own task or dataset: (give details below)

To reproduce

It is about documentation, just look at it.

Expected behavior

[Narsil]

Hi @zhaowei-wang98 ,

The documentation is sort of correct I think.

The values will be defaulted from the config values of the model, which themselves have a default value which is the one provided in the docs: check here how those variables are initialized https://github.com/huggingface/transformers/blob/master/src/transformers/generation_utils.py#L1062

So even though, technically they are None by default, they will actually inherit the default of the config values (which can be overidden too).
It seems max_length does say the default values comes from the config, where other values don't necessarily.
@patrickvonplaten Should we harmonize?

So I think the doc reflects reality as much as possible. https://huggingface.co/docs/transformers/v4.17.0/en/main_classes/model#transformers.generation_utils.GenerationMixin.generate

But maybe some value is incorrectly reported in particular ?
Did you encounter a particular issue with the default values ?

[patrickvonplaten]

We do write this in the docs:

Apart from inputs, all the arguments below will default to the value of the attribute of the same name inside the PretrainedConfig of the model. The default values indicated are the default values of those config.

Let's maybe wrap this message in a warning to make it clearer? Putting it in my generate doc refactor PR: #15988

[github-actions]

This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.

Please note that issues that do not follow the contributing guidelines are likely to be ignored.