search

google / ml-metadata

For recording and retrieving metadata associated with ML developer and data scientist workflows.
https://www.tensorflow.org/tfx/guide/mlmd
Apache License 2.0
626 stars 148 forks source link

When supports for attrs version >21 is available? #159

Closed amirhessam88 closed 2 years ago

amirhessam88 commented 2 years ago

Currently, ml-metadata (1.9.0) depends on attrs (>=20.3,<21) and this raises a lot of incompatibility of other wrappers on top of ml-metadata. I am wondering when the support for attrs > 21 is available?

BrianSong commented 2 years ago

Hi @amirhessam88, could you be more specific about the wrappers on top of ml-metadata that causing the incompatibility? That could help us be more clear about the situation.

amirhessam88 commented 2 years ago

Hey @BrianSong ; some growing libraries such as ZenML depends on ml-pipelines-sdk and accordingly that depends on ml-metadata; at the same time, if another projects requires attrs >=21 that would create a dependency conflict. does that make sense?

BrianSong commented 2 years ago

I see.

However, ml_metadata has the version constraint in order to be sync with TFX as shown in [1].

Let me talk to the TFX folks to see if they can bump that version.

[1] https://github.com/tensorflow/tfx/blob/b021a486c9363147eca35e97742b555a5ad1cce4/tfx/dependencies.py#L77

amirhessam88 commented 2 years ago

@BrianSong Sounds good. Thanks

BrianSong commented 2 years ago

Hi, @amirhessam88, I bump the limit to <22 so that at least 21.X attrs version can be supported. The reason why I cannot bump the version higher is because our internal attrs only support <22 and we need to keep that in sync.

amirhessam88 commented 2 years ago

Hey @BrianSong ; I totally understand the reasonings and really appreciate it. 🙏

BrianSong commented 2 years ago

Closing this issue right now and please feel free to reopen it if this does not work for you.