Fall back to Blackwell for legacy manifests missing gpu_architecture#21
Merged
Merged
Conversation
Same bug class as SparkDistill's dataset_verify fix: verify_manifest_policy required gpu_architecture to resolve from either the manifest's own field or gpu_profile.gpu_architecture — both only populated by the post-Hopper build_manifest_v2/require_supported_gpu. Bundles published before this field existed have neither, only the older gpu_profile.family, so verification started rejecting every pre-existing Blackwell bundle with "gpu_architecture must be one of [...], got None". Now falls back to "blackwell" when gpu_architecture is absent from both locations but gpu_profile.family == "blackwell" — the only architecture that existed before this field was added, so this can't admit anything actually unsupported.
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
Companion fix to SparkDistill#106 — same bug class, discovered while chasing why the dataset registry's mining-dataset aggregation started failing today.
verify_manifest_policyrequiredgpu_architectureto resolve from either the manifest's own field orgpu_profile.gpu_architecture— both only populated by the post-Hopperbuild_manifest_v2/require_supported_gpu. Bundles published before this field existed have neither (only the oldergpu_profile.family), sosparkproof-verifystarted rejecting every pre-existing Blackwell bundle withgpu_architecture must be one of [...], got None.Fix: fall back to
"blackwell"when both lookups come up empty butgpu_profile.family == "blackwell"— the only architecture that existed before this field was added, so this can't admit anything actually unsupported.Test plan
uv run pytest -q— 281 passed, 9 skipped