v0.15.0
What's Changed
New Features
- Document autocompletion enablement by @scaliby in #796
- Remove default --no-enable-autoupgrade for GPUs not using CT by @jamOne- in #794
- Telemetry client id generation by @scaliby in #802
- feat: Update Kueue to v0.14.3 by @jamOne- in #804
- Validate CPU and memory limits against the machine type. by @SikaGrr in #808
- Add MetricsCollector by @scaliby in #809
- Server telemetry reporting by @scaliby in #814
- Add support for shared TPU reservations by @SikaGrr in #823
- Flush to clearcut synchronously if uploader script cannot be found by @scaliby in #829
- Enrich telemetry metadata by @scaliby in #830
- Add ability to disable telemetry through config by @scaliby in #831
- Log cluster create arguments by @scaliby in #833
- Add --sub-slicing to cluster create-pathways and set default for create-ray by @jamOne- in #837
- Fix sub-slicing topology and level labels by @jamOne- in #836
- Display tmp_file content in --dry-run by @jamOne- in #844
Bug fixes
- Fix dws cluster delete by @scaliby in #788
- Fix dependency in nightly by @scaliby in #805
- Revert "Kill python3 xpk.py" by @SikaGrr in #821
- Fix single-host TPU nodepools, always set num-nodes if not flex. by @SikaGrr in #822
- Fix credentials retrieval by @scaliby in #826
- Workload policy label by @scaliby in #817
- Prevent the use of workload policy for single host nodepools by @FIoannides in #825
- Add missing valid TPU topologies for v4, v5p and tpu7x. by @SikaGrr in #834
- If multiple topologies map to the same number of cores, prefer default one. by @SikaGrr in #838
- Fix e2e tests by @FIoannides in #842
- Add always to DWS cluster delete by @scaliby in #849
Full Changelog: v0.14.3...v0.15.0