forked from opendatahub-io/kserve
-
Notifications
You must be signed in to change notification settings - Fork 1
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
[Cherry Pick] Multi node implementation (#434)
* Multi-Node Inference Implementation (kserve#3972) Signed-off-by: jooho lee <[email protected]> * fix lint and unit test for odh Signed-off-by: jooho lee <[email protected]> --------- Signed-off-by: jooho lee <[email protected]>
- Loading branch information
Showing
43 changed files
with
8,394 additions
and
348 deletions.
There are no files selected for viewing
This file contains 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 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 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
180 changes: 180 additions & 0 deletions
180
config/runtimes/kserve-huggingfaceserver-multinode.yaml
This file contains 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 +1,180 @@ | ||
| apiVersion: serving.kserve.io/v1alpha1 | ||
| kind: ClusterServingRuntime | ||
| metadata: | ||
| name: kserve-huggingfaceserver-multinode | ||
| spec: | ||
| annotations: | ||
| prometheus.kserve.io/port: "8080" | ||
| prometheus.kserve.io/path: "/metrics" | ||
| supportedModelFormats: | ||
| - name: huggingface | ||
| version: "1" | ||
| autoSelect: true | ||
| priority: 2 | ||
| protocolVersions: | ||
| - v2 | ||
| - v1 | ||
| containers: | ||
| - name: kserve-container | ||
| image: kserve/huggingfaceserver:latest | ||
| command: ["bash", "-c"] | ||
| args: | ||
| - | | ||
| ray start --head --disable-usage-stats --include-dashboard false | ||
| # wait for other node to join | ||
| until [[ $(ray status | grep -c node_) -eq ${PIPELINE_PARALLEL_SIZE} ]]; do | ||
| echo "Waiting..." | ||
| sleep 1 | ||
| done | ||
| ray status | ||
| export MODEL=${MODEL_ID} | ||
| if [[ ! -z ${MODEL_DIR} ]] | ||
| then | ||
| MODEL=${MODEL_DIR} | ||
| fi | ||
| python3 -m huggingfaceserver --model_name=${MODEL_NAME} --model_dir=${MODEL} --tensor-parallel-size=${TENSOR_PARALLEL_SIZE} --pipeline-parallel-size=${PIPELINE_PARALLEL_SIZE} | ||
| env: | ||
| - name: RAY_PORT | ||
| value: "6379" | ||
| - name: RAY_ADDRESS | ||
| value: 127.0.0.1:6379 | ||
| - name: POD_NAMESPACE | ||
| valueFrom: | ||
| fieldRef: | ||
| fieldPath: metadata.namespace | ||
| - name: VLLM_CONFIG_ROOT | ||
| value: /tmp | ||
| - name: HF_HUB_CACHE | ||
| value: /tmp | ||
| resources: | ||
| requests: | ||
| cpu: "2" | ||
| memory: 6Gi | ||
| limits: | ||
| cpu: "4" | ||
| memory: 12Gi | ||
| volumeMounts: | ||
| - name: shm | ||
| mountPath: /dev/shm | ||
| livenessProbe: | ||
| failureThreshold: 3 | ||
| periodSeconds: 30 | ||
| successThreshold: 1 | ||
| timeoutSeconds: 5 | ||
| initialDelaySeconds: 10 | ||
| exec: | ||
| command: | ||
| - bash | ||
| - -c | ||
| - | | ||
| ./huggingfaceserver/health_check.py liveness | ||
| readinessProbe: | ||
| failureThreshold: 2 | ||
| periodSeconds: 10 | ||
| successThreshold: 1 | ||
| timeoutSeconds: 5 | ||
| initialDelaySeconds: 10 | ||
| exec: | ||
| command: | ||
| - bash | ||
| - -c | ||
| - | | ||
| ./huggingfaceserver/health_check.py readiness ${PIPELINE_PARALLEL_SIZE} http://localhost:8080 | ||
| startupProbe: | ||
| failureThreshold: 40 | ||
| periodSeconds: 30 | ||
| successThreshold: 1 | ||
| timeoutSeconds: 5 | ||
| initialDelaySeconds: 5 | ||
| exec: | ||
| command: | ||
| - bash | ||
| - -c | ||
| - | | ||
| ./huggingfaceserver/health_check.py startup | ||
| volumes: | ||
| - name: shm | ||
| emptyDir: | ||
| medium: Memory | ||
| sizeLimit: 3Gi | ||
| workerSpec: | ||
| pipelineParallelSize: 2 | ||
| tensorParallelSize: 1 | ||
| containers: | ||
| - name: worker-container | ||
| image: kserve/huggingfaceserver:latest | ||
| command: ["bash", "-c"] | ||
| args: | ||
| - | | ||
| SECONDS=0 | ||
| while true; do | ||
| if (( SECONDS <= 120 )); then | ||
| if ray health-check --address "${HEAD_SVC}.${POD_NAMESPACE}.svc.cluster.local:6379" > /dev/null 2>&1; then | ||
| echo "Global Control Service(GCS) is ready." | ||
| break | ||
| fi | ||
| echo "$SECONDS seconds elapsed: Waiting for Global Control Service(GCS) to be ready." | ||
| else | ||
| if ray health-check --address "${HEAD_SVC}.${POD_NAMESPACE}.svc.cluster.local:6379"; then | ||
| echo "Global Control Service(GCS) is ready. Any error messages above can be safely ignored." | ||
| break | ||
| fi | ||
| echo "$SECONDS seconds elapsed: Still waiting for Global Control Service(GCS) to be ready." | ||
| fi | ||
| sleep 5 | ||
| done | ||
| RAY_HEAD_ADDRESS="${HEAD_SVC}.${POD_NAMESPACE}.svc.cluster.local:6379" | ||
| echo "Attempting to connect to Ray cluster at $RAY_HEAD_ADDRESS ..." | ||
| ray start --address="$RAY_HEAD_ADDRESS" --block | ||
| env: | ||
| - name: POD_NAME | ||
| valueFrom: | ||
| fieldRef: | ||
| fieldPath: metadata.name | ||
| - name: POD_NAMESPACE | ||
| valueFrom: | ||
| fieldRef: | ||
| fieldPath: metadata.namespace | ||
| resources: | ||
| requests: | ||
| cpu: "2" | ||
| memory: 6Gi | ||
| limits: | ||
| cpu: "4" | ||
| memory: 12Gi | ||
| volumeMounts: | ||
| - name: shm | ||
| mountPath: /dev/shm | ||
| livenessProbe: | ||
| failureThreshold: 3 | ||
| periodSeconds: 5 | ||
| successThreshold: 1 | ||
| timeoutSeconds: 5 | ||
| exec: | ||
| command: | ||
| - bash | ||
| - -c | ||
| - | | ||
| ./huggingfaceserver/health_check.py registered_nodes ${PIPELINE_PARALLEL_SIZE} | ||
| startupProbe: | ||
| failureThreshold: 12 | ||
| periodSeconds: 5 | ||
| successThreshold: 1 | ||
| timeoutSeconds: 5 | ||
| exec: | ||
| command: | ||
| - bash | ||
| - -c | ||
| - | | ||
| ./huggingfaceserver/health_check.py startup | ||
| volumes: | ||
| - name: shm | ||
| emptyDir: | ||
| medium: Memory | ||
| sizeLimit: 3Gi |
This file contains 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 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
Some generated files are not rendered by default. Learn more about how customized files appear on GitHub.
Oops, something went wrong.
This file contains 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
Oops, something went wrong.