Skip to content

Conversation

@arikaufman
Copy link

This draft introduces modifications to support batching. The batching stack will be developed in parallel with the regular stack. Batched changes will be released incrementally, layer by layer, to allow easier review.

@tekaratzas, I’m thinking we could review these changes layer by layer. Once we’re confident in the full batching flow, we can swap out the main flow with the batched version.

Do we want to:

  1. Create separate PRs for each layer and review/merge them one by one, or
  2. Create one large PR in a separate subdirectory, then perform the swap in a follow-up PR?

@arikaufman arikaufman changed the title Output_Projection Batching Draft: Add Support for Batching Oct 12, 2025
@arikaufman
Copy link
Author

Output Projection Batching
The changes required for output projection batching are now complete. These changes are in development under the batch_mode directory. As of now, I plan to continue updating this PR with support for each layer, as I progress.

@arikaufman
Copy link
Author

Feed Forward Batching
The changes required for the feed forward layer of batching are now complete (tests included).

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant