rename(release): the CUDA archive is cuda, not cuda-fat (#117) - #411
Merged
Conversation
The release tuples were `linux-x86_64-glibc-cuda-fat` and `linux-aarch64-glibc-cuda-fat`. "fat" is a build-time property of the gencode, not something a user downloading a CUDA build needs in the filename -- the metadata map already recorded the backend as plain `cuda`, so the suffix was describing the same thing twice and only in the artifact name. 33 tuple occurrences across 13 files, plus the two other artifact-facing spots: the `_vllm_server_artifact_backend` value in examples/CMakeLists.txt and the prose archive name in docs/USAGE.md. DELIBERATELY NOT RENAMED: `check-cuda-fat-gencode.py`, the `cuda-fat-build` CI job, the `build-cuda-fat` build directory and the `cuda-fat-*` Triton AOT test cases. Those name the FAT GENCODE property -- that one binary carries every supported SM -- which is a real thing the checker verifies and which stays true regardless of what the archive is called. Renaming them would lose the distinction the audit exists to make. Safe to do now and awkward later: docs/STATUS.md still records "no published binaries", so nothing external depends on these names. After the first publication this would be a breaking change to download URLs. Gates: preflight rc=0; test_release_manifest, test_release_pipeline, test_release_accelerator_metadata, test_release_archive, test_release_index, test_release_metadata, test_check_release_binary_contract and test_cpu_release_gates all green; check-release-workflow OK. The manifest fixtures under tests/scripts/fixtures/release_manifest/v1/ are renamed with the tuples they pin, so the contract is checked against the new names rather than carrying the old ones forward. Issue: #117 FOLLOWING_AGENTS_PROTOCOL Following-Agents-Protocol: true AI-Assisted: true Assisted-by: AGENT:claude-opus-5 [Claude Code]
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
2 participants
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.
Issue #117 (
ENG-RELEASE-BINARIES).The release tuples were
linux-x86_64-glibc-cuda-fat/linux-aarch64-glibc-cuda-fat. "fat" is a build-time property of the gencode, not something a user downloading a CUDA build needs in the filename — andrelease_accelerator_metadata.pyalready recorded the backend as plaincuda, so the suffix said the same thing twice, in the one place users see.Now
linux-x86_64-glibc-cuda/linux-aarch64-glibc-cuda.What changed
33 tuple occurrences across 13 files, plus the two other artifact-facing spots:
_vllm_server_artifact_backendinexamples/CMakeLists.txt, and the prose archive name indocs/USAGE.md.What deliberately did NOT change
check-cuda-fat-gencode.py, thecuda-fat-buildCI job, thebuild-cuda-fatdirectory, and thecuda-fat-*Triton AOT test cases.Those name the fat gencode property — that one binary carries every supported SM — which is a real thing the checker verifies and stays true regardless of what the archive is called. Renaming them would erase the distinction the audit exists to make.
Timing
Safe now, awkward later:
docs/STATUS.mdstill records "no published binaries", so nothing external depends on these names. After first publication this becomes a breaking change to download URLs.Evidence
preflight rc=0. All eight release suites green —
test_release_manifest,test_release_pipeline,test_release_accelerator_metadata,test_release_archive,test_release_index,test_release_metadata,test_check_release_binary_contract,test_cpu_release_gates— pluscheck-release-workflowOK.The manifest fixtures under
tests/scripts/fixtures/release_manifest/v1/are renamed along with the tuples they pin, so the contract is verified against the new names rather than carrying the old ones forward.🤖 Generated with Claude Code