[PowerX] verify NVIDIA SRT completion and failure evidence / 验证 NVIDIA SRT 完成状态及失败证据 - #3052

Draft

edwingao28 wants to merge 17 commits into

mainfrom

fix/powerx-slurm-completion-split

Commits on Sep 12, 2026

fix: verify Slurm completion and exit llm-d workers normally
edwingao28
committed
85986f4
fix: parse Slurm allocation exit codes without derived status
edwingao28
committed
86b8621
ci: run llm-d lifecycle regressions
edwingao28
committed
20e5e71
chore: preserve Slurm diagnostics after current main sync
edwingao28
committed
87b9e45
test: synchronize shared replay signal readiness
edwingao28
committed
fe3a0d4
chore: merge main diagnostics into Slurm lifecycle fixes
edwingao28
committed
a620774
fix: recheck llm-d completion after engine shutdown
edwingao28
committed
8f89342
chore: retain native power foundation in Slurm fixes
edwingao28
committed
8742b8b

Commits on Sep 13, 2026

chore: sync Slurm completion branch with main
edwingao28
committed
cf2f442
chore: sync Slurm completion with TileRT main updates
edwingao28
committed
db219ab
fix: retain Slurm outcome with GB200 AgentX power collection
edwingao28
committed
32bdb7a
fix: preserve GB300 native status when syncing main
edwingao28
committed
7d330f8
fix: preserve H200 native failure through power collection
edwingao28
committed
8d7d93c
fix: retain B300 recovery in Slurm completion refresh
edwingao28
committed
6d4b874
fix: sync Slurm completion with required H100 power
edwingao28
committed
eaa2f4b

Commits on Sep 14, 2026

fix: verify NVIDIA SRT allocation completion
edwingao28
committed
25cf855
fix: preserve failed H200 signal status
edwingao28
committed
728ad5a