Fix wrong documentation of ignore_unused_parameters

[UniverseFly]

The documentation says:

ignore_unused_parameters: bool = True
...
This is set to False (which should be True) by default, which means unused parameters are ignored and training continues.

Hence the inconsistency.

[UniverseFly]

Hi @tjruwase, the original commit encountered some unexpected unit test errors even if it only modified the docstring. Could you kindly check if there's anything wrong? Thanks.

[loadams]

Hi @tjruwase, the original commit encountered some unexpected unit test errors even if it only modified the docstring. Could you kindly check if there's anything wrong? Thanks.

Hi @UniverseFly - there is a known CI issues we are working on. There doesn't appear to be anything wrong with your PR, we will get it merged shortly, thanks!

[keunwoochoi]

i think we still see the inconsistent documentation on this..

[loadams]

Hi @keunwoochoi - could you link the inconsistent docs? We'd certainly like to fix them if there is anything inconsistent

[keunwoochoi]

hi @loadams, yes absolutely. on the doc about the json (https://www.deepspeed.ai/docs/config-json/), the "Description" column says False is the default behavior but the "Default" column specifies it's True.

[loadams]

@keunwoochoi - the docs inconsistency should be fixed here: #4949

Link to the code where the default is set: https://github.com/microsoft/DeepSpeed/blob/13d84b4912492d40b68f4da7bc570e47b54bce2a/deepspeed/runtime/zero/config.py#L242C4-L242C42

[keunwoochoi]

@loadams thanks!