[Feature]: Alternating local-global attention layers

[griff4692]

🚀 The feature, motivation and pitch

Gemma-2 and new Ministral models use alternating sliding window and full attention layers to reduce the size of the KV cache.

The KV cache is a huge inference bottleneck and this technique could be fine-tuned into other models to make them much more memory efficient, especially for large batch sizes.

Alternatives

No response

Additional context

No response

Before submitting a new issue...

• [X] Make sure you already searched for relevant issues, and asked the chatbot living at the bottom right corner of the documentation page, which can answer lots of frequently asked questions.

[simon-mo]

Yes. This will be worked on, added to the roadmap.

[github-actions[bot]]

This issue has been automatically marked as stale because it has not had any activity within 90 days. It will be automatically closed if no further activity occurs within 30 days. Leave a comment if you feel this issue should remain open. Thank you!

[github-actions[bot]]

This issue has been automatically closed due to inactivity. Please feel free to reopen if you feel it is still relevant. Thank you!