Record bounded DFlash7 Python-overlay live validation - #138
Closed
FujitsuPolycom wants to merge 2 commits into
Closed
Record bounded DFlash7 Python-overlay live validation#138FujitsuPolycom wants to merge 2 commits into
FujitsuPolycom wants to merge 2 commits into
Conversation
This was referenced Aug 30, 2026
Owner
Author
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Resulting behavior
Adds a bounded live-validation record for GLM-5.3 DFlash7 image
sha256:9faa36a9f37aee16d97ab9214ef3153b4d200121126e6b2dee5ebb63109fea18, whose OCI revision label ise2d92fdc7d0306d664d6fd9f296dc2adcaf0fe05and whose SparkCache source is commit5d571018de5b63a9a90e5c11e6d6e86bbff4a957.The record qualifies only the observed four-rank startup and health gates, raw-completion smoke, exact flat restore, reconstructed page-delta correctness, exact 128K-class persistent restore, and one retained-prefix wave each at concurrency 2, 8, and 16. Page-delta performance is research-only. DFlash response quality is unsupported by this evidence.
The executable quickstart documents the exact canonical-to-legacy SparkCache key translation required by this image. It also identifies SparkRing pull request #137 as an unbuilt image contract that uses canonical CUDA keys directly.
Compatibility and cache namespace
No cache identity, digest salt, chunk geometry, or publication schema changes. Existing cache entries retain their namespaces. The record preserves the
tail-cow-v1topage-tail-cow-v1publication contract and documents known geometry and null-block limitations without widening qualification.Validation
python -m pytest performance/records/glm53-flash/test_dflash7_pr25_live_validation.py scripts/test_prepare_glm53_dflash7_python_overlay_profile.py runtime/glm53-flash-dflash7-python-overlay/test_dflash7_python_overlay.py -q— 19 passedpython -m ruff check performance/records/glm53-flash/test_dflash7_pr25_live_validation.py— passedgit diff --check— passedpython -m pytest -q— 1,979 passed, 9 skipped, 1 pre-existing unrelated failure inscripts/test_glm53_flash_profile.py::test_public_glm53_benchmark_is_sanitized_and_bounded; the test expects root README wording absent from the stacked baseNo image was built or published. No serving process or host configuration was changed.