GQA Attention

[RishikeshMagar]

Implemented GQA attention with the tests. Discussed with buttercutter and Leo Chen on Discord

Issue: #52

[pascalnotin]

Thank you @RishikeshMagar -- looks good to me! One suggestion before we merge in main: we should add a parameter to the config file that allows to select the type of attention we want in the APT instance (standard vs grouped-query) and adapt the forward method of the APTAttention class accordingly.

[RishikeshMagar]

Hey @pascalnotin,
Sorry about the delay in response. I know the project has gone through a reboot and I have been following the progress discord.

I have made the changes and it is on my fork but I am not entirely sure how the requested changes work, do I submit a new PR or just keep the same one.

[pascalnotin]

Thanks for making the change @RishikeshMagar, and apologies for the delayed review! Ok to now merge this PR into main.