* Separate validation logic out into a ValidatingAdmissionWebhook * Add gradient accumulation * AdaScale Improvements * Add BERT examples * Add methods to output tensorboard metrics * Using jsonpatch and podpreset to accept cluster config * Add job trigger with synchronization
0.2.4
* Introduce non-preemptible jobs through `preemptible` flag in job spec. * Support `minReplicas` job parameter for both non-preemptible and preemptible jobs.
0.2.3
Maintenance release
- Allocator bugfix - Change helm chart name to adaptdl-sched
0.2.2
Changes: * Push docker images through local mitmproxy to remote insecure registry * Use port 32000 for registry service everywhere to gain compatibility with microk8s * Handle case where registry is deployed in a different namespace * Update readme and copyrights
0.2.1
Changes:
* Changes the adaptdl scheduler docker image name from `adaptdl` to `adaptdl-sched`