perf(llama32_1b): dual-hart K-split - #221
Open
karabambus wants to merge 1 commit into
Open
Conversation
Q8_0 decode + track claim
github-actions
Bot
added
track: week-2-challenge
karabambus
marked this pull request as draft
July 26, 2026 08:25
karabambus
marked this pull request as ready for review
July 26, 2026 08:25
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Changes
The N=1 decode dot product for Q8_0 previously gave each hart its own independent output rows. This change splits along K instead: the two sibling harts of a minion share a row and each computes half of the reduction. One partial is published through the shire-local L2 scratchpad and handed over with an FCC semaphore handshake; hart 0 sums the two halves and stores the result.
Results
Board measurements,
llama-serverdecode, Llama-3.2-1B-Instruct Q8_0:Recipe
Unfortunately I didn't have opportunity to properly work on this, I studied #159 and made some improvements and implemented it on llama and that is basically it.