Bind DFlash7 to canonical SparkCache CUDA keys - #137
Closed
FujitsuPolycom wants to merge 1 commit into
Closed
Conversation
Advance the exact DFlash7 image and both loader profiles to SparkCache commit b95aa8ab, Git tree 723fc604, and independently reproduced deployable-source digest 48e008ba. Prepared image labels and verification now require the canonical-v1 CUDA configuration schema. Profiles use only spark_cache_cuda_restore, CUDA placement library, digest, arena, and restore-worker keys. No PR25 compatibility profile or legacy-key rewrite is required by the launcher path. Cache compatibility: checkpoint identities, page-tail publication schema, record vocabulary, digest salts, TP/DCP geometry, vLLM patch bytes, lease contract, and CUDA placement ABI are unchanged. Compatible PR25 page-tail entries remain in the same namespace. Validation: 29 focused tests passed; exact context preparation reproduced the PR26 commit, tree, source digest, runtime patch receipt, and canonical image labels; maintained suite reported 1,964 passed and 9 skipped with one unrelated stacked-base README assertion failure; Ruff, Python, JSON, Bash, diff, and prose checks passed.
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Result
Advances the exact DFlash7 image and both loader profiles from SparkCache PR #25 to PR #26 canonical CUDA configuration.
The image pins SparkCache commit
b95aa8ab0068dc66a6892a5c311d7e9dd4a9c55a, Git tree723fc604d73911a9e907798fd7932e4fc9c95df5, and independently reproduced clean-source SHA-25648e008ba0cbd12f1ffae1c28388ea83310f41c6219c955e13d63ab171290d8de. Prepared image labels and the verifier requireorg.sparkcache.cuda-config-schema=canonical-v1.Configuration
Both DFlash7 profiles use only the canonical SparkCache CUDA restore, placement-library, digest, arena, and restore-worker keys. The current launcher can consume either profile directly; no PR25 compatibility profile or legacy-key rewrite is required.
The DFlash draft-loader patch, vLLM native/Python identities, B12X, checkpoint identities, DFlash depth, draft TP4, page-tail behavior, and vLLM ownership contract remain unchanged.
Cache compatibility
No cache namespace change. Checkpoint identities, publication schema, record vocabulary, digest salts, TP/DCP geometry, vLLM patch bytes, lease contract, and CUDA placement ABI are unchanged. Compatible PR25 page-tail entries remain eligible.
Validation
No image was built or published, and no service was modified.