Repository navigation
docs: design hierarchical external KV cache - #1
Merged
Merged
Conversation
There was a problem hiding this comment.
Copilot was unable to review this pull request because the user who requested the review has insufficient remaining quota. Try requesting a review at a lower effort level.
An organization owner can choose the repository organization as the billing source for reviews requested by members with a paid Copilot license. Learn about Copilot code review billing.
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
This proposal extends the existing prefix-storage owners into a hierarchical external KV interface. OrbitInfer keeps scheduling, prefix matching and device-page leases; a selected provider owns DRAM/SSD/peer residency. The first integration is OrbitKV, followed by separate Mooncake Store and LMCache adapters.
The design separates logical outcome, borrowed-buffer retirement and cache publication; requires representation identity and bounded byte/operation accounting; and retains the existing
lookup/load/storeboundary. It includes failure controls and phased implementation. No adapter or runtime behavior is implemented here.The parallelism guide now distinguishes imported TP/attention-DP/CP/EP/expert-TP source from unimplemented PP and encoder/prefill/decode disaggregation, and from installed multi-GPU qualification.
Validation on the local Apple Silicon development environment:
uv run --no-sync python scripts/check.py: passed (format, Clippy, 310 Rust tests, doctests, Ruff, 2,405 Python tests passed / 551 skipped, installed CPU example, core and model wheel boundaries).python3 scripts/check_source.py: all 616 imported files and provenance retained.git diff --check: passed.The first main CI run already has unrelated failures at the base commit: new stable Clippy diagnostics, Linux missing tilelang during collection, and macOS missing torchvision/model package import in child processes. Apple Silicon tests and lint/types passed. The plan records clean-environment CI as a prerequisite to provider implementation; this documentation-only PR does not fix or conceal those failures.
Handoff:
10ef0ee; baseline:200cac4.01d9e36c; official Mooncake Store / LMCache interface documentation checked 2026-11-15.