-
Notifications
You must be signed in to change notification settings - Fork 83
Add Gemma 4 vLLM runtimes #599
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Open
ankrovv
wants to merge
5
commits into
ome-projects:main
Choose a base branch
from
ankrovv:feat/gemma4
base: main
Could not load branches
Branch not found: {{ refName }}
Loading
Could not load tags
Nothing to show
Loading
Are you sure you want to change the base?
Some commits from the old base branch may be removed from the timeline,
and old review comments may become outdated.
Open
Changes from all commits
Commits
File filter
Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
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 |
|---|---|---|
| @@ -0,0 +1,16 @@ | ||
| apiVersion: ome.io/v1beta1 | ||
| kind: ClusterBaseModel | ||
| metadata: | ||
| name: gemma-4-26b-a4b-it | ||
| spec: | ||
| modelCapabilities: | ||
| - TEXT_TO_TEXT | ||
| - IMAGE_TEXT_TO_TEXT | ||
| - VIDEO_TEXT_TO_TEXT | ||
| vendor: google | ||
| disabled: false | ||
| version: "1.0.0" | ||
| displayName: google.gemma-4-26b-a4b-it | ||
| storage: | ||
| storageUri: hf://google/gemma-4-26B-A4B-it | ||
| path: /raid/models/google/gemma-4-26B-A4B-it |
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 +1,16 @@ | ||
| apiVersion: ome.io/v1beta1 | ||
| kind: ClusterBaseModel | ||
| metadata: | ||
| name: gemma-4-31b-it | ||
| spec: | ||
| modelCapabilities: | ||
| - TEXT_TO_TEXT | ||
| - IMAGE_TEXT_TO_TEXT | ||
| - VIDEO_TEXT_TO_TEXT | ||
| vendor: google | ||
| disabled: false | ||
| version: "1.0.0" | ||
| displayName: google.gemma-4-31b-it | ||
| storage: | ||
| storageUri: hf://google/gemma-4-31B-it | ||
| path: /raid/models/google/gemma-4-31B-it |
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 +1,17 @@ | ||
| apiVersion: ome.io/v1beta1 | ||
| kind: ClusterBaseModel | ||
| metadata: | ||
| name: gemma-4-e2b-it | ||
| spec: | ||
| modelCapabilities: | ||
| - TEXT_TO_TEXT | ||
| - IMAGE_TEXT_TO_TEXT | ||
| - VIDEO_TEXT_TO_TEXT | ||
| - AUDIO_TEXT_TO_TEXT | ||
| vendor: google | ||
| disabled: false | ||
| version: "1.0.0" | ||
| displayName: google.gemma-4-e2b-it | ||
| storage: | ||
| storageUri: hf://google/gemma-4-E2B-it | ||
| path: /raid/models/google/gemma-4-E2B-it |
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 +1,17 @@ | ||
| apiVersion: ome.io/v1beta1 | ||
| kind: ClusterBaseModel | ||
| metadata: | ||
| name: gemma-4-e4b-it | ||
| spec: | ||
| modelCapabilities: | ||
| - TEXT_TO_TEXT | ||
| - IMAGE_TEXT_TO_TEXT | ||
| - VIDEO_TEXT_TO_TEXT | ||
| - AUDIO_TEXT_TO_TEXT | ||
| vendor: google | ||
| disabled: false | ||
| version: "1.0.0" | ||
| displayName: google.gemma-4-e4b-it | ||
| storage: | ||
| storageUri: hf://google/gemma-4-E4B-it | ||
| path: /raid/models/google/gemma-4-E4B-it |
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
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 +1,256 @@ | ||
| apiVersion: ome.io/v1beta1 | ||
| kind: ClusterServingRuntime | ||
| metadata: | ||
| name: vllm-gemma-4-tp1 | ||
| spec: | ||
| disabled: false | ||
| acceleratorRequirements: | ||
| acceleratorClasses: | ||
| - nvidia-h100-1 | ||
| - nvidia-h100-2 | ||
| - nvidia-h100-4 | ||
| - nvidia-h100-8 | ||
| - nvidia-a100-80gb-1 | ||
| - nvidia-a100-80gb-2 | ||
| - nvidia-a100-80gb-4 | ||
| - nvidia-a100-80gb-8 | ||
| - nvidia-h200-1 | ||
| - nvidia-h200-2 | ||
| - nvidia-h200-4 | ||
| - nvidia-h200-8 | ||
| - nvidia-b200-1 | ||
| - nvidia-b200-2 | ||
| - nvidia-b200-4 | ||
| - nvidia-b200-8 | ||
| supportedModelFormats: | ||
| - modelFramework: | ||
| name: transformers | ||
| version: "5.5.0.dev0" | ||
| modelFormat: | ||
| name: safetensors | ||
| version: "1.0.0" | ||
| modelArchitecture: Gemma4ForConditionalGeneration | ||
| autoSelect: true | ||
| priority: 1 | ||
| version: "1.0.0" | ||
| acceleratorConfig: | ||
| nvidia-h100-1: | ||
| tensorParallelismOverride: | ||
| tensorParallelSize: 1 | ||
| nvidia-h100-2: | ||
| tensorParallelismOverride: | ||
| tensorParallelSize: 2 | ||
| nvidia-h100-4: | ||
| tensorParallelismOverride: | ||
| tensorParallelSize: 4 | ||
| nvidia-h100-8: | ||
| tensorParallelismOverride: | ||
| tensorParallelSize: 8 | ||
| nvidia-a100-80gb-1: | ||
| tensorParallelismOverride: | ||
| tensorParallelSize: 1 | ||
| nvidia-a100-80gb-2: | ||
| tensorParallelismOverride: | ||
| tensorParallelSize: 2 | ||
| nvidia-a100-80gb-4: | ||
| tensorParallelismOverride: | ||
| tensorParallelSize: 4 | ||
| nvidia-a100-80gb-8: | ||
| tensorParallelismOverride: | ||
| tensorParallelSize: 8 | ||
| nvidia-h200-1: | ||
| tensorParallelismOverride: | ||
| tensorParallelSize: 1 | ||
| nvidia-h200-2: | ||
| tensorParallelismOverride: | ||
| tensorParallelSize: 2 | ||
| nvidia-h200-4: | ||
| tensorParallelismOverride: | ||
| tensorParallelSize: 4 | ||
| nvidia-h200-8: | ||
| tensorParallelismOverride: | ||
| tensorParallelSize: 8 | ||
| nvidia-b200-1: | ||
| tensorParallelismOverride: | ||
| tensorParallelSize: 1 | ||
| nvidia-b200-2: | ||
| tensorParallelismOverride: | ||
| tensorParallelSize: 2 | ||
| nvidia-b200-4: | ||
| tensorParallelismOverride: | ||
| tensorParallelSize: 4 | ||
| nvidia-b200-8: | ||
| tensorParallelismOverride: | ||
| tensorParallelSize: 8 | ||
| modelSizeRange: | ||
| min: 4.6B | ||
| max: 27.7B | ||
| protocolVersions: | ||
| - openAI | ||
| routerConfig: | ||
| annotations: | ||
| prometheus.io/scrape: "true" | ||
| prometheus.io/port: "29000" | ||
| prometheus.io/path: "/metrics" | ||
| labels: | ||
| logging-forward: enabled | ||
| runner: | ||
| name: router | ||
| image: docker.io/lightseekorg/smg:1.4.1 | ||
| ports: | ||
| - containerPort: 8080 | ||
| name: http | ||
| resources: | ||
| limits: | ||
| cpu: "1" | ||
|
ankrovv marked this conversation as resolved.
|
||
| memory: 2Gi | ||
| args: | ||
| - --host | ||
| - 0.0.0.0 | ||
| - --port | ||
| - "8080" | ||
| - --service-discovery | ||
| - --service-discovery-namespace | ||
| - $(NAMESPACE) | ||
| - --service-discovery-port | ||
| - "8080" | ||
| - --selector | ||
| - component=engine ome.io/inferenceservice=$(INFERENCESERVICE_NAME) | ||
| - --enable-igw | ||
| - --request-id-headers | ||
| - opc-request-id | ||
| - --log-json | ||
| - --disable-retries | ||
| - --disable-circuit-breaker | ||
| - --disable-tokenizer-autoload | ||
| env: | ||
| - name: NAMESPACE | ||
| valueFrom: | ||
| fieldRef: | ||
| fieldPath: metadata.namespace | ||
| - name: INFERENCESERVICE_NAME | ||
| valueFrom: | ||
| fieldRef: | ||
| fieldPath: metadata.labels['ome.io/inferenceservice'] | ||
| readinessProbe: | ||
| httpGet: | ||
| path: /readiness | ||
| port: 8080 | ||
| failureThreshold: 5 | ||
| periodSeconds: 30 | ||
| timeoutSeconds: 10 | ||
| livenessProbe: | ||
| httpGet: | ||
| path: /liveness | ||
| port: 8080 | ||
| failureThreshold: 5 | ||
| periodSeconds: 30 | ||
| timeoutSeconds: 10 | ||
| startupProbe: | ||
| httpGet: | ||
| path: /health | ||
| port: 8080 | ||
| failureThreshold: 10 | ||
| periodSeconds: 20 | ||
| timeoutSeconds: 10 | ||
| engineConfig: | ||
| annotations: | ||
| prometheus.io/scrape: "true" | ||
| prometheus.io/port: "8080" | ||
| prometheus.io/path: "/metrics" | ||
| labels: | ||
| logging-forward: enabled | ||
| tolerations: | ||
| - key: "nvidia.com/gpu" | ||
| operator: "Exists" | ||
| effect: "NoSchedule" | ||
| volumes: | ||
| - name: dshm | ||
| emptyDir: | ||
| medium: Memory | ||
| affinity: | ||
| nodeAffinity: | ||
| requiredDuringSchedulingIgnoredDuringExecution: | ||
| nodeSelectorTerms: | ||
| - matchExpressions: | ||
| - key: node.kubernetes.io/instance-type | ||
| operator: In | ||
| values: | ||
| - BM.GPU.A100-v2.8 | ||
| - BM.GPU.H100.8 | ||
| - BM.GPU.H200-NC.8 | ||
| - BM.GPU.H200.8 | ||
| runner: | ||
| name: ome-container | ||
| image: fra.ocir.io/idqj093njucb/official-vllm-openai:v0.19.1-nightly-gemma4 | ||
| ports: | ||
| - containerPort: 8080 | ||
| name: http1 | ||
| protocol: TCP | ||
| command: | ||
| - /bin/bash | ||
| - '-lc' | ||
| - -- | ||
| args: | ||
| - | | ||
| vllm serve \ | ||
| --port=8080 \ | ||
| --model="$MODEL_PATH" \ | ||
| --max-log-len=0 \ | ||
| --served-model-name=vllm-model \ | ||
| --tensor-parallel-size=1 \ | ||
| --max-model-len=-1 \ | ||
| --gpu-memory-utilization=0.9 \ | ||
| --enable-auto-tool-choice \ | ||
| --tool-call-parser=gemma4 \ | ||
| --reasoning-parser=gemma4 \ | ||
| --async-scheduling \ | ||
| --no-scheduler-reserve-full-isl \ | ||
| --limit-mm-per-prompt '{"image": 10, "audio": 1, "video": 1}' | ||
| env: | ||
| - name: VLLM_LOGGING_LEVEL | ||
| value: "INFO" | ||
| - name: VLLM_RPC_TIMEOUT | ||
| value: '30000' | ||
| - name: VLLM_ENGINE_ITERATION_TIMEOUT_S | ||
| value: '120' | ||
| volumeMounts: | ||
| - mountPath: /dev/shm | ||
| name: dshm | ||
| resources: | ||
| requests: | ||
| cpu: 10 | ||
| memory: 80Gi | ||
| nvidia.com/gpu: 1 | ||
| limits: | ||
| cpu: 10 | ||
| memory: 80Gi | ||
| nvidia.com/gpu: 1 | ||
|
|
||
| readinessProbe: | ||
| httpGet: | ||
| path: /health | ||
| port: 8080 | ||
| failureThreshold: 3 | ||
| successThreshold: 1 | ||
| periodSeconds: 60 | ||
| timeoutSeconds: 200 | ||
|
|
||
| livenessProbe: | ||
| httpGet: | ||
| path: /health | ||
| port: 8080 | ||
| failureThreshold: 5 | ||
| successThreshold: 1 | ||
| periodSeconds: 60 | ||
| timeoutSeconds: 60 | ||
|
|
||
| startupProbe: | ||
| httpGet: | ||
| path: /health | ||
| port: 8080 | ||
| failureThreshold: 150 | ||
| successThreshold: 1 | ||
| periodSeconds: 6 | ||
| initialDelaySeconds: 60 | ||
| timeoutSeconds: 30 | ||
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.
The change didn't follow old pattern to create one runtime for each model, why we switch to use one runtime for multiple models?
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.
All the Gemma4 models use the same underlying model architecture so creating one runtime per model would create redundancies and consolidating by tp-size also allows for future expansion and support if google were to release new gemma4 models using the same architecture. cc @YouNeedCryDear for more context
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.
Correct. We are not going direction of one runtime per model as it scaling. If multiple models in the same family are sharing the same architecture, same format and essentially same engine config. Then we are combining those into a single runtime. Parallelism and engine args overwrite will be controlled on Accelerator Class level. Please let me know if there is any concerns for it @XinyueZhang369