Repository navigation
Conversation
There was a problem hiding this comment.
I reviewed this PR and didn't find any bugs. The changed paths have CODEOWNERS (the AMD config and recipe owners, plus core for the changelog), and the run still needs an accepted /use <run_id>, so a human look would still be worthwhile.
What was reviewed:
- The image bump for
dsv4-fp4-mi355x-sglang-agentic-mtpinconfigs/amd-master.yaml. The newrocm10tag uses the same naming as another MI355X sglang entry already in that file. - The
enable-amd-monokernel: trueline, which is added only tooverride_tp8_c1and sits alongside that override's other sglang serving flags. - The new
perf-changelog.yamlblock, which is appended at the tail with existing bytes untouched and has one entry for the affected key. Itspr-linkis still thepull/XXXplaceholder, as several earlier entries also are.
Extended reasoning...
This changes three YAML config lines for one benchmark key: a container image bump (rocm720 to rocm10), a new sglang monokernel flag at TP8/C1, and an appended perf-changelog entry. It touches no auth, injection or data-exposure surface. I deferred instead of approving because CODEOWNERS covers amd-master.yaml, the mi355x-* srt-slurm recipes and the changelog. The changelog entry also triggers result collection, which needs a human-accepted run before merge.
This review covers commit 6757fb8, which is no longer the latest commit on this pull request; later commits are not covered by it.
No description provided.