[CI] Rename cuda dimension to platform in anyscale CUDA build steps#62274
[CI] Rename cuda dimension to platform in anyscale CUDA build steps#62274andrew-anyscale wants to merge 1 commit intoandrew/revup/master/array-wheel-buildfrom
Conversation
|
Reviews in this chain: |
|
There was a problem hiding this comment.
Code Review
This pull request refactors the Buildkite and Docker configurations by replacing the cuda matrix dimension with a platform dimension, ensuring that CUDA versions are explicitly prefixed (e.g., cu12.3.2). These changes are applied across several build steps, including ray-anyscale-cuda-build, ray-llm-anyscale-cuda-build, and ray-ml-anyscale-cuda-build. Feedback indicates that for full consistency and to meet the PR's stated goals, the anyscalemlbuild step should also be updated to include the platform dimension.
35916c4 to
c927698
Compare
aa202ec to
8fd435a
Compare
c927698 to
0a16bfc
Compare
8fd435a to
b1e601c
Compare
483f20d to
794e740
Compare
290fda6 to
98f547b
Compare
794e740 to
a0d4a61
Compare
98f547b to
b80e07e
Compare
a0d4a61 to
b1d1bf4
Compare
b80e07e to
eacf845
Compare
b1d1bf4 to
3e39fed
Compare
3e39fed to
6c31f0e
Compare
eacf845 to
14dfb8f
Compare
Rename the cuda matrix dimension to platform (with cu-prefixed values) in the three anyscale CUDA build steps and their wandas. The wandas now accept $PLATFORM directly instead of constructing it from cu$CUDA_VERSION; output is identical since callers pass cu-prefixed values. This aligns the dimension name with the publish steps (anyscalebuild, anyscalellmbuild, anyscalemlbuild) which already use a platform dimension, enabling precise ($) dependency matching on both platform and python when these steps are later converted to array syntax. Topic: wanda-platform-env Relative: array-wheel-build Signed-off-by: andrew <[email protected]>
14dfb8f to
f966b57
Compare
6c31f0e to
ce67dac
Compare
Rename the cuda matrix dimension to platform (with cu-prefixed values)
in the three anyscale CUDA build steps and their wandas. The wandas
now accept $PLATFORM directly instead of constructing it from
cu$CUDA_VERSION; output is identical since callers pass cu-prefixed
values.
This aligns the dimension name with the publish steps (anyscalebuild,
anyscalellmbuild, anyscalemlbuild) which already use a platform
dimension, enabling precise ($) dependency matching on both platform
and python when these steps are later converted to array syntax.
Topic: wanda-platform-env
Relative: array-wheel-build
Signed-off-by: andrew [email protected]