Revert max_size from 1GB to 128MB to fix KV cache regression#2737
Draft
AMD-yanfeiwang wants to merge 1 commit intoROCm:mainfrom
Draft
Revert max_size from 1GB to 128MB to fix KV cache regression#2737AMD-yanfeiwang wants to merge 1 commit intoROCm:mainfrom
AMD-yanfeiwang wants to merge 1 commit intoROCm:mainfrom
Conversation
The max_size change in 8cfe5e2 (1024*1024*1024 = 1GB) causes: 1. Prefill allreduce dispatch change: QuickReduce INT8 -> custom_all_reduce 2. GPU memory increase: +3.6GB/GPU for allreduce buffers 3. KV cache reduction: -75K tokens (-1.68%), -2.47GB This causes MTP accept rate regression at C>=2 (-8~10%) and throughput regression (-4~10%) while only C=1 improves slightly. Reverting to 8192*1024*8*2 = 128MB restores the old dispatch behavior and recovers full KV cache capacity.
Contributor
🏷️ CI GuideRuns automatically on every PR:
Extended tests (opt-in via labels):
|
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
The max_size change in 8cfe5e2 (102410241024 = 1GB) causes:
This causes MTP accept rate regression at C>=2 (-8
10%) and throughput regression (-410%) while only C=1 improves slightly.Reverting to 819210248*2 = 128MB restores the old dispatch behavior and recovers full KV cache capacity.
Motivation
Technical Details
Test Plan
Test Result
Submission Checklist