Correct histogram publication ordering in the MTP3 mesh image - #226
Closed
FujitsuPolycom wants to merge 2 commits into
Closed
Correct histogram publication ordering in the MTP3 mesh image#226FujitsuPolycom wants to merge 2 commits into
FujitsuPolycom wants to merge 2 commits into
Conversation
The derived mesh image checks the pinned selector preimage, adds the block-entry barrier, and increments the kernel compile revision. Compute receipts and package maps identify the transformed source while retaining transport and native-library identities. Local patch tests and parent/result image-content verification pass; GPU validation is separate.
Owner
Author
|
Status: implemented in SparkRing main through merged #236. The histogram publication-ordering correction in this PR is incorporated; this PR is closed as superseded by that integration. The component head The native-MTP3 cache/checkpoint guide selects the integrated image and deployment inputs. Closing this component PR does not broaden the profile's research-only hardware qualification. |
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Status: research-only serving image with bounded GPU validation. Related to #224; extends the compute/transport composition in #219.
The derived image adds block-wide synchronization before B12X histogram publication and increments the compiled-kernel revision. The installer checks the exact mesh selector preimage and updates the compute package map, source lock, embedded profile, and source receipt. It preserves the parent transport bundle, CUDA components, vLLM overrides, marker, and native libraries. The unmodified mesh verifier passes before and after installation.
Published image:
ghcr.io/fujitsupolycom/sparkring-glm53-sparkcache@sha256:7aa57ed1901b83c8581c53cb233db930bb1a133d74be01168d967b9dd56b3e49Validation: 348 local tests passed with three platform skips. On GB10, the mesh parent helper exposed incomplete publication in 20/20 probes; the entry-synchronized helper passed 20/20. The built child passed fifteen GPU correctness cases and 2,000 graph replays with exact selected-index parity, contexts up to 200K tokens, and concurrent copies. External image verification and anonymous pull passed.
No serving containers were replaced. No full-model startup, TP4 collective test, actual SparkCache restore, or soak was run with this child. The managed renderer retains its parent-image pins; deployment requires an explicit source-bound profile update and coordinated validation. The running SparkCache telemetry composition is outside this artifact. The README and validation receipt identify image, parent, source lock, cache namespace, test conditions, and limitations.