Repository navigation

Issues

New issue

Kubeflow Trainer v2.4 Release
#4188 · andreyvelich opened on Oct 7, 2026
6
KEP-3562: OptimizationJob CRD for HPO: The Evolution of the Katib Project
#3562 · akshaychitneni opened on Jan 27, 2026
26
KEP-3015: Support Workload API for TrainJob Scheduling
#3015 · andreyvelich opened on Dec 2, 2025
14

Proposal: allow selecting ReplicatedJobs for PodGroupPolicy in sequential runtimes
kubeflow/trainer#4298 · opened on Oct 23, 2026
TrainJob accepts env names and resource requests that its Pods cannot use, and then never starts
kubeflow/trainer#4292 · opened on Oct 21, 2026
OptimizationJob accepts trainer env vars its trials can't use, then stalls without failing
kubeflow/trainer#4285 · opened on Oct 21, 2026
1
bug(runtime): MPI and DeepSpeed worker nodes do not receive resourcesPerNode and custom Image from TrainJob
kubeflow/trainer#4283 · opened on Oct 20, 2026
1
perf(cache): reuse one Flight connection per worker instead of reconnecting for every batch
kubeflow/trainer#4280 · opened on Oct 20, 2026
1
DRA: resolve numProcPerNode from ResourceClaimTemplate (KEP-2782 follow-up)
kubeflow/trainer#4267 · opened on Oct 19, 2026
1
[BUG]: Data cache discards failures and reports success
kubeflow/trainer#4261 · opened on Oct 18, 2026
1
TrainJob accepts names that are too long for the JobSet it creates, and the TrainJob then never starts
kubeflow/trainer#4260 · opened on Oct 18, 2026
11
[PERF]: Data cache does quadratic work at startup and on every query
kubeflow/trainer#4258 · opened on Oct 17, 2026
2
bug: RuntimePatch manager validation loses list index in error field path
kubeflow/trainer#4256 · opened on Oct 17, 2026
Quick survey: your experience as a PyTorch Ecosystem maintainer
kubeflow/trainer#4252 · opened on Oct 16, 2026
1
bug(runtime): Runtime resource creation errors not surfaced in TrainJob status
kubeflow/trainer#4251 · opened on Oct 16, 2026
1
E2E: test the MPI runtimes with Open MPI 4
kubeflow/trainer#4241 · opened on Oct 14, 2026
2
CPU E2E: local-container-mnist notebook intermittently hangs because the node-1 container never exits
kubeflow/trainer#4240 · opened on Oct 14, 2026
3
bug(runtime): prevent existing TrainJob runtime snapshots from being overwritten
kubeflow/trainer#4238 · opened on Oct 14, 2026
3
Document the custom metrics and the scraping configuration
kubeflow/trainer#4215 · opened on Oct 12, 2026
5
Expose custom trainer metrics via the /metrics scrape endpoint
kubeflow/trainer#4214 · opened on Oct 12, 2026
5
Add ServiceMonitor to Helm and Kustomize deploy configuration.
kubeflow/trainer#4213 · opened on Oct 12, 2026
3
bug(controller): TrainJob reconciler reverts external JobSet suspension and omits JobSetSuspended from status
kubeflow/trainer#4212 · opened on Oct 12, 2026
7
MPI runtime numNodes scales the launcher Job instead of the node Job
kubeflow/trainer#4210 · opened on Oct 12, 2026
3
bug: Flux plugin passes per-node GPU count to -g flag instead of per-task count
kubeflow/trainer#4206 · opened on Oct 9, 2026
bug: Flux shared-memory volume mounted read-only, preventing writes to /dev/shm
kubeflow/trainer#4204 · opened on Oct 9, 2026
Deleting a TrainingRuntime can leave its TrainJobs stuck in Terminating
kubeflow/trainer#4203 · opened on Oct 9, 2026
1
Torch, JAX and XGBoost coordinator address ignores the JobSet network.subdomain
kubeflow/trainer#4198 · opened on Oct 8, 2026
1
Helm chart cannot enable metrics auth or TLS options: ClusterRole and config template have drifted from the generated manifests
kubeflow/trainer#4196 · opened on Oct 7, 2026
1