Conversation
Add fine sub-block projection and carry external chunk geometry from the vLLM adapter into engine-driven SHM and pickle transfers. Preserve coarse identity behavior and fail before slot reservation on invalid geometry. Co-authored-by: Hermes Agent <hermes-agent@nousresearch.com> Signed-off-by: myshytf <9619163+myshytf@users.noreply.github.com>
|
Important Review skippedAuto reviews are disabled on base/target branches other than the default branch. 🗂️ Base branches to auto review (1)
Please check the settings in the CodeRabbit UI or the ⚙️ Run configurationConfiguration used: Organization UI Review profile: CHILL Plan: Pro Plus Run ID: You can disable this status message by setting the Use the checkbox below for a quick retry:
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
Avoid applying a whole-dict fused-attention layout detector to mixed rank-4 attention and rank-3 Mamba tensors. Preserve the legacy physical-block check for homogeneous and single-group registrations while delegating multigroup validation to the existing per-group engine-driven context. Signed-off-by: myshytf <9619163+myshytf@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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
Why
Kimi-K3 DCP8 uses a 12,288-token logical manager block while LMCache needs 1,536-token external objects. The release branch requires both the generic projection core and vLLM adapter metadata; either half alone is not functional.
Validation
Source-matched
NO_GPU_EXT=1run:tests/v1/multiprocess/test_dcp_fine_external_chunks.pytests/v1/test_lmcache_mp_connector_fine_chunks.pytests/v1/test_kv_cache_groups.pytests/v1/test_vllm_mp_adapter.pyAlso passed focused
ruff check,ruff format --check, static security scan, andgit diff --check.Compatibility
Coarse chunks retain identity block IDs and the pre-existing skip behavior. Fine projection activates only when explicit group geometry proves an exact integer subdivision. Invalid or incomplete geometry fails before SHM slot reservation.