Validate vLLM P/D cache-source metrics on GB300 - #2797

Open

cquil11 wants to merge 20 commits into

mainfrom

codex/h100-minimaxm3-disagg-source

Commits on Sep 1, 2026

Add H100 MiniMax-M3 disaggregated source validation
cquil11
committed
b3f8ba4
Schedule H100 disaggregated source validation
cquil11
committed
cf67f7d
Allow cold H100 disaggregated container startup
cquil11
committed
572b087
Revert "Allow cold H100 disaggregated container startup"
cquil11
committed
0bb3a24
Revert "Schedule H100 disaggregated source validation"
cquil11
committed
1f50088
Revert "Add H100 MiniMax-M3 disaggregated source validation"
cquil11
committed
4308bee
Use proven GB300 disaggregated validation point
cquil11
committed
c221f79
Match proven GB300 c256 topology
cquil11
committed
66e7bf4
Pin validation image in GB300 recipe
cquil11
committed
3057a73
fix: align disaggregated validation with upstream MoE backend
cquil11
committed
cb93ced
fix: collect worker metrics for custom AgentX validation
cquil11
committed
7e709c0
Merge upstream main to refresh validation sweep
cquil11
committed
600b1dc
fix(agentx): isolate P/D profiling metrics from warmup
cquil11
committed
edc96a6
fix: retain Mooncake store checkpoints in the P/D validation image
cquil11
committed
ef79a0b

Commits on Sep 2, 2026

Merge upstream main before corrected P/D validation
cquil11
committed
4567fc0

Commits on Sep 3, 2026