What's Changed
- [Doc] Bring the README up to date with how Plumb behaves now by @bluayer in #16
- [Feature] Choose the adaptive path's mode and chooser per policy by @bluayer in #17
- [Feature] Burst limits, load trend and early capacity for the adaptive path by @bluayer in #18
- [Bugfix] Close gaps in the adaptive path's early capacity by @bluayer in #19
- [Doc] Explain the HPA scale-down window after Plumb lowers a floor by @bluayer in #20
- [Feature] Leave pending replicas the HPA holds out of a member's shortage by @bluayer in #21
- [Bugfix] Move traffic only for relief, and bind planner plans to their state by @bluayer in #22
- [Refactor] Name node pools and launch failures without Karpenter or AWS terms by @bluayer in #23
- [Bugfix] Don't borrow for pods the scheduler has already placed by preemption by @bluayer in #24
- [Refactor] Split large files, move model hosts to internal/model, drop global state by @bluayer in #25
- [Refactor] Split the hub step into phases and test what only e2e covered by @bluayer in #26
- [CI] Check the Helm chart on every PR and install it in kind by @bluayer in #27
- [Feature] Validate chart values with a schema and spread replicas over nodes by @bluayer in #28
- [Bugfix] Don't borrow while the home cluster's own nodes are launching by @bluayer in #29
- [Doc] Show results on real GPUs under Why Plumb by @bluayer in #30
Full Changelog: v0.1.0-rc.2...v0.1.0