[None][fix] Batch addSequence with pre-claim to fix host offloading M…#12878
Open
liji-nv wants to merge 2 commits intoNVIDIA:feat/bench_yfrom
Open
[None][fix] Batch addSequence with pre-claim to fix host offloading M…#12878liji-nv wants to merge 2 commits intoNVIDIA:feat/bench_yfrom
liji-nv wants to merge 2 commits intoNVIDIA:feat/bench_yfrom
Conversation
liji-nv
force-pushed
the
fix/batch-addsequence-mnt-overflow
branch
3 times, most recently
from
ba22dfe to
cecae98
Compare
…NT overflow When host offloading is enabled, onboarding a host block to GPU during addSequence can trigger eviction of other reusable host blocks from the radix tree. This causes actual KV cache reuse to be less than the scheduler estimated, leading to max_num_tokens (MNT) overflow assertions. Add a new addSequenceBatch API that processes all first-chunk context requests in two phases: - Phase 1: Walk the radix tree and claimBlock() for all matching blocks across all requests. No onboarding, no allocation. This protects reusable blocks from eviction. - Phase 2: Onboard host blocks and allocate non-matching blocks. Since all reusable blocks are already claimed, evictions during onboarding cannot touch them. On the Python side, replace the TOCTOU-prone revalidation loop (count_reusable_blocks + budget check) with a single batch call. Signed-off-by: Jin Li <59594262+liji-nv@users.noreply.github.com>
liji-nv
force-pushed
the
fix/batch-addsequence-mnt-overflow
branch
from
cecae98 to
9dc7da9
Compare
…chContent (NVIDIA#12550) Signed-off-by: Aurelien Chartier <2567591+achartier@users.noreply.github.com>
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
2 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
…NT overflow
When host offloading is enabled, onboarding a host block to GPU during addSequence can trigger eviction of other reusable host blocks from the radix tree. This causes actual KV cache reuse to be less than the scheduler estimated, leading to max_num_tokens (MNT) overflow assertions.
Add a new addSequenceBatch API that processes all first-chunk context requests in two phases:
On the Python side, replace the TOCTOU-prone revalidation loop (count_reusable_blocks + budget check) with a single batch call.
@coderabbitai summary
Description
Test Coverage
PR Checklist
Please review the following before submitting your PR:
PR description clearly explains what and why. If using CodeRabbit's summary, please make sure it makes sense.
PR Follows TRT-LLM CODING GUIDELINES to the best of your knowledge.
Test cases are provided for new code paths (see test instructions)
Any new dependencies have been scanned for license and vulnerabilities
CODEOWNERS updated if ownership changes
Documentation updated as needed
Update tava architecture diagram if there is a significant design change in PR.
The reviewers assigned automatically/manually are appropriate for the PR.
Please check this after reviewing the above items as appropriate for this PR.
GitHub Bot Help
To see a list of available CI bot commands, please comment
/bot help.