Feature/attn decode dynamic sched#46
Open
shaochangxu wants to merge 2 commits into
Open
Conversation
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Online inference decode phase suffers from significant differences in KV cache lengths across requests in the same batch.Existing attention operators partition tasks based on a static grid dimension, this causes long requests to become bottlenecks or short requests to generate many empty tasks, leading to severe CTA load imbalance.We implement a dynamic task scheduling scheme: the KV caches of all requests are first divided into uniform tile granules (64 tokens). A greedy bin-packing algorithm then distributes these tiles evenly across all CTAs, making the workload of each CTA nearly equal and fundamentally eliminating the long-tail effect.
