Integrate grouped-query attention into APT class

[pascalnotin]

1. We should have one parameter in the config that specifies the type of attention ('standard' vs 'grouped-query')
2. We should extend the APTAttention class to include grouped-query attention
3. The PR should include unit test to confirm grouped-query attention works as expected

[RishikeshMagar]

Hi, I'm new here and would like to contribute to this project. Can I take a look at this issue? I'm still getting the hang of unit tests, so I might need some guidance in that area. Thanks!

[pascalnotin]

Hi @RishikeshMagar -- welcome to the project! Two other colleagues (Buttercutter and Leo Chen on discord) have expressed interest in this issue as well and have started coordinating on an approach. They will be reaching out to you re: next steps / how to collaborate on this issue :)

[Leo-T-Zang]

/take

[Leo-T-Zang]

Hi @RishikeshMagar, What is your discord username? We can discuss there.

[RishikeshMagar]

Thanks @pascalnotin. I will connect with them.

Hi @Leo-T-Zang,
Sure, we can connect on discord. My discord user name is Rishi. Thanks!

[Leo-T-Zang]

@RishikeshMagar, can you send a messege to me? user name is Leo Chen and handle is tz6352

[RishikeshMagar]

Sure @Leo-T-Zang, I will send you a message on discord sometime today.

[RishikeshMagar]

@pascalnotin
Just an FYI, I have implemented this and butter cutter from discord is helping me out with tests for this. I should be able to submit a PR this weekend

[pascalnotin]

Fantastic - thank you both!