integration: validate selected-record CKV transports on latest Sparkinfer#4
Draft
FujitsuPolycom wants to merge 6 commits into
Draft
Conversation
Add a direct CUDA-IPC exchange for destination-selected fixed-width records, expose it through the Sparkinfer PCIe facade, and preserve runtime-JIT CUDA sources in built wheels. Co-authored-by: OpenAI Codex <codex@openai.com> Signed-off-by: FujitsuPolycom <87842395+FujitsuPolycom@users.noreply.github.com>
Add single-plane and three-plane CUDA-IPC copy-engine exchange with rotated peer scheduling, bounded slab overlays, graph-safe deferred release, and large-offset coverage. Co-authored-by: OpenAI Codex <codex@openai.com> Signed-off-by: FujitsuPolycom <87842395+FujitsuPolycom@users.noreply.github.com>
Assisted-by: OpenAI Codex Signed-off-by: FujitsuPolycom <87842395+FujitsuPolycom@users.noreply.github.com>
FujitsuPolycom
force-pushed
the
codex/sparse-ckv-integration-20260721
branch
from
July 21, 2026 22:56
77a75e7 to
31d296c
Compare
Owner
Author
|
@coderabbitai review |
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Purpose
Fork-only integration gate for selected-record CKV transport and bounded
long-context indexer scratch, rebased onto
local-inference-lab/sparkinfermasteratbce2c2b.No upstream PR is authorized by this integration PR.
What is included
The paired vLLM integration is
FujitsuPolycom/vllm#7.
Long-context scratch fix
The paged-indexer planner keeps the fast two-level fold when its candidate-plane
scratch fits the configured cap. Larger requests use the existing bounded
two-buffer streaming carry fold instead of allocating scratch proportional to
the complete prompt.
The final model completed clean cold-prefill gates at:
This fixes the prior 163,840-token OOM without reducing the 304,863-token KV
pool or changing short-context routing.
Validation
Scope
behavior
Credits
The selected-record design is primarily based on Koush's sparse-CKV and MTP
deduplication work, with Luke Alonso's B12X/Sparkinfer infrastructure.
Implementation, testing, benchmarking, and documentation were developed with
OpenAI Codex assistance and validated on the target host.