feat(runtimes): add support for ClusterTrainingRuntimes in Helm chart #3124
There are no files selected for viewing
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
|
|
@@ -31,6 +31,57 @@ Alternatively, you can install the latest version from the master branch (e.g. ` | |
| helm install kubeflow-trainer oci://ghcr.io/kubeflow/charts/kubeflow-trainer --version 0.0.0-sha-bfccb7b | ||
| ``` | ||
|
|
||
| ### Install with ClusterTrainingRuntimes | ||
|
|
||
| You can optionally deploy ClusterTrainingRuntimes as part of the Helm installation. Runtimes are disabled by default to keep the chart lightweight. | ||
|
|
||
| To enable specific runtimes: | ||
|
|
||
| ```bash | ||
| helm install kubeflow-trainer oci://ghcr.io/kubeflow/charts/kubeflow-trainer \ | ||
| --version 2.1.0 \ | ||
| --set runtimes.torchDistributed.enabled=true \ | ||
| --set runtimes.deepspeedDistributed.enabled=true | ||
| ``` | ||
|
|
||
| Or use a custom values file: | ||
|
|
||
| ```yaml | ||
| # values.yaml | ||
| runtimes: | ||
| torchDistributed: | ||
| enabled: true | ||
| torchDistributedWithCache: | ||
| enabled: true | ||
| dataCache: | ||
| enabled: true | ||
| cacheImage: | ||
| tag: "v2.0.0" | ||
| deepspeedDistributed: | ||
| enabled: true | ||
| mlxDistributed: | ||
| enabled: true | ||
|
|
||
| # Required for torch-distributed-with-cache | ||
| dataCache: | ||
| enabled: true | ||
| ``` | ||
|
|
||
| Then install with: | ||
|
|
||
| ```bash | ||
| helm install kubeflow-trainer oci://ghcr.io/kubeflow/charts/kubeflow-trainer \ | ||
| --version 2.1.0 \ | ||
| -f values.yaml | ||
| ``` | ||
|
|
||
| ### Available Runtimes | ||
|
Member
There was a problem hiding this comment. Please can you also:
|
||
|
|
||
| - **torch-distributed**: PyTorch distributed training (no custom images) | ||
| - **torch-distributed-with-cache**: PyTorch with distributed data cache support (requires `dataCache.enabled=true`) | ||
| - **deepspeed-distributed**: DeepSpeed distributed training with MPI | ||
| - **mlx-distributed**: MLX distributed training with MPI | ||
|
|
||
| ### Uninstall the chart | ||
|
|
||
| ```shell | ||
|
|
@@ -72,6 +123,30 @@ See [helm uninstall](https://helm.sh/docs/helm/helm_uninstall) for command docum | |
| | dataCache.enabled | bool | `false` | Enable/disable data cache support (LWS dependency, ClusterRole). Set to `true` to install data cache components. | | ||
| | dataCache.lws.install | bool | `true` | Whether to install LeaderWorkerSet as a dependency. Set to `false` if LeaderWorkerSet is already installed in the cluster. | | ||
| | dataCache.lws.fullnameOverride | string | `"lws"` | String to fully override LeaderWorkerSet release name. | | ||
| | runtimes | object | `{"deepspeedDistributed":{"enabled":false,"image":{"registry":"ghcr.io","repository":"kubeflow/trainer/deepspeed-runtime","tag":""}},"defaultEnabled":false,"mlxDistributed":{"enabled":false,"image":{"registry":"ghcr.io","repository":"kubeflow/trainer/mlx-runtime","tag":""}},"torchDistributed":{"enabled":false},"torchDistributedWithCache":{"dataCache":{"cacheImage":{"registry":"ghcr.io","repository":"kubeflow/trainer/data-cache","tag":""},"enabled":true},"enabled":false},"torchtuneDistributed":{"enabled":false,"image":{"registry":"ghcr.io","repository":"kubeflow/trainer/torchtune-runtime","tag":""}}}` | ClusterTrainingRuntimes configuration These are optional runtime templates that can be deployed with the Helm chart. Each runtime provides a blueprint for different ML frameworks and configurations. | | ||
| | runtimes.defaultEnabled | bool | `false` | Enable all default runtimes (torch, deepspeed, mlx, torchtune) when set to true. Individual runtime settings will be ignored if this is enabled. | | ||
| | runtimes.torchDistributed | object | `{"enabled":false}` | PyTorch distributed training runtime (no custom images required) | | ||
| | runtimes.torchDistributed.enabled | bool | `false` | Enable deployment of torch-distributed runtime | | ||
| | runtimes.torchDistributedWithCache | object | `{"dataCache":{"cacheImage":{"registry":"ghcr.io","repository":"kubeflow/trainer/data-cache","tag":""},"enabled":true},"enabled":false}` | PyTorch distributed training with data cache support | | ||
| | runtimes.torchDistributedWithCache.enabled | bool | `false` | Enable deployment of torch-distributed-with-cache runtime | | ||
| | runtimes.torchDistributedWithCache.dataCache.cacheImage.registry | string | `"ghcr.io"` | Data cache image registry | | ||
| | runtimes.torchDistributedWithCache.dataCache.cacheImage.repository | string | `"kubeflow/trainer/data-cache"` | Data cache image repository | | ||
| | runtimes.torchDistributedWithCache.dataCache.cacheImage.tag | string | `""` | Data cache image tag. Defaults to chart version if empty. | | ||
| | runtimes.deepspeedDistributed | object | `{"enabled":false,"image":{"registry":"ghcr.io","repository":"kubeflow/trainer/deepspeed-runtime","tag":""}}` | DeepSpeed distributed training runtime | | ||
| | runtimes.deepspeedDistributed.enabled | bool | `false` | Enable deployment of deepspeed-distributed runtime | | ||
| | runtimes.deepspeedDistributed.image.registry | string | `"ghcr.io"` | DeepSpeed runtime image registry | | ||
| | runtimes.deepspeedDistributed.image.repository | string | `"kubeflow/trainer/deepspeed-runtime"` | DeepSpeed runtime image repository | | ||
| | runtimes.deepspeedDistributed.image.tag | string | `""` | DeepSpeed runtime image tag. Defaults to chart version if empty. | | ||
| | runtimes.mlxDistributed | object | `{"enabled":false,"image":{"registry":"ghcr.io","repository":"kubeflow/trainer/mlx-runtime","tag":""}}` | MLX distributed training runtime | | ||
| | runtimes.mlxDistributed.enabled | bool | `false` | Enable deployment of mlx-distributed runtime | | ||
| | runtimes.mlxDistributed.image.registry | string | `"ghcr.io"` | MLX runtime image registry | | ||
| | runtimes.mlxDistributed.image.repository | string | `"kubeflow/trainer/mlx-runtime"` | MLX runtime image repository | | ||
| | runtimes.mlxDistributed.image.tag | string | `""` | MLX runtime image tag. Defaults to chart version if empty. | | ||
| | runtimes.torchtuneDistributed | object | `{"enabled":false,"image":{"registry":"ghcr.io","repository":"kubeflow/trainer/torchtune-runtime","tag":""}}` | TorchTune distributed training runtime | | ||
| | runtimes.torchtuneDistributed.enabled | bool | `false` | Enable deployment of torchtune-distributed runtime | | ||
| | runtimes.torchtuneDistributed.image.registry | string | `"ghcr.io"` | TorchTune runtime image registry | | ||
| | runtimes.torchtuneDistributed.image.repository | string | `"kubeflow/trainer/torchtune-runtime"` | TorchTune runtime image repository | | ||
| | runtimes.torchtuneDistributed.image.tag | string | `""` | TorchTune runtime image tag. Defaults to chart version if empty. | | ||
|
|
||
| ## Maintainers | ||
|
|
||
|
|
||
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,35 @@ | ||
| # Example values.yaml configuration for deploying ClusterTrainingRuntimes | ||
|
Member
There was a problem hiding this comment. Let's remove this example from this PR, I think it is sufficient to explain what is needed in the README. |
||
|
|
||
| # Deploy torch-distributed runtime (no custom images needed) | ||
| runtimes: | ||
| torchDistributed: | ||
| enabled: true | ||
|
|
||
| # Deploy torch-distributed with data cache support | ||
| torchDistributedWithCache: | ||
| enabled: true | ||
| # cacheImage will use chart version by default | ||
| # To override, specify custom tag: | ||
| # cacheImage: | ||
| # tag: "v2.0.0" | ||
|
|
||
| # Deploy DeepSpeed runtime | ||
khushiiagrawal marked this conversation as resolved.
Show resolved
Hide resolved
|
||
| deepspeedDistributed: | ||
| enabled: true | ||
| # Override image tag if needed: | ||
| # image: | ||
| # tag: "custom-v1.0.0" | ||
|
|
||
| # Deploy MLX runtime | ||
| mlxDistributed: | ||
| enabled: false | ||
| # Can enable and customize: | ||
| # enabled: true | ||
| # image: | ||
| # registry: my-registry.io | ||
| # repository: custom/mlx-runtime | ||
| # tag: "v1.0.0" | ||
|
|
||
| # Note: torch-distributed-with-cache requires data cache support | ||
| dataCache: | ||
| enabled: true | ||
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,75 @@ | ||
| {{- /* | ||
| Copyright 2025 The Kubeflow authors. | ||
|
|
||
| Licensed under the Apache License, Version 2.0 (the "License"); | ||
| you may not use this file except in compliance with the License. | ||
| You may obtain a copy of the License at | ||
|
|
||
| https://www.apache.org/licenses/LICENSE-2.0 | ||
|
|
||
| Unless required by applicable law or agreed to in writing, software | ||
| distributed under the License is distributed on an "AS IS" BASIS, | ||
| WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | ||
| See the License for the specific language governing permissions and | ||
| limitations under the License. | ||
| */ -}} | ||
|
|
||
| {{- if .Values.runtimes.deepspeedDistributed.enabled }} | ||
|
||
| apiVersion: trainer.kubeflow.org/v1alpha1 | ||
| kind: ClusterTrainingRuntime | ||
| metadata: | ||
| name: deepspeed-distributed | ||
| labels: | ||
| trainer.kubeflow.org/framework: deepspeed | ||
| {{- include "trainer.labels" . | nindent 4 }} | ||
| spec: | ||
| mlPolicy: | ||
| numNodes: 1 | ||
| mpi: | ||
| numProcPerNode: 1 | ||
| mpiImplementation: OpenMPI | ||
| sshAuthMountPath: /home/mpiuser/.ssh | ||
| runLauncherAsNode: true | ||
| template: | ||
| spec: | ||
| network: | ||
| publishNotReadyAddresses: true | ||
| successPolicy: | ||
| operator: All | ||
| targetReplicatedJobs: | ||
| - launcher | ||
| replicatedJobs: | ||
| - name: launcher | ||
| template: | ||
| metadata: | ||
| labels: | ||
| trainer.kubeflow.org/trainjob-ancestor-step: trainer | ||
| spec: | ||
| template: | ||
| spec: | ||
| containers: | ||
| - name: node | ||
| image: {{ include "trainer.runtimeImage" (list .Values.runtimes.deepspeedDistributed.image .) }} | ||
| securityContext: | ||
| runAsUser: 1000 | ||
| - name: node | ||
| template: | ||
| spec: | ||
| template: | ||
| spec: | ||
| containers: | ||
| - name: node | ||
| image: {{ include "trainer.runtimeImage" (list .Values.runtimes.deepspeedDistributed.image .) }} | ||
| securityContext: | ||
| runAsUser: 1000 | ||
| command: | ||
| - /usr/sbin/sshd | ||
| args: | ||
| - -De | ||
| - -f | ||
| - /home/mpiuser/.sshd_config | ||
| readinessProbe: | ||
| tcpSocket: | ||
| port: 2222 | ||
| initialDelaySeconds: 5 | ||
| {{- end }} | ||
Oops, something went wrong.
Oops, something went wrong.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Add example where enabledDefault is true