Add vLLM inference engine support (#398)

Signed-off-by: Dolpher Du <[email protected]>
Co-authored-by: Gang G Li <[email protected]>

helm-charts/common/vllm/.helmignore

@@ -0,0 +1,23 @@
+# Patterns to ignore when building packages.
+# This supports shell glob matching, relative path matching, and
+# negation (prefixed with !). Only one pattern per line.
+.DS_Store
+# Common VCS dirs
+.git/
+.gitignore
+.bzr/
+.bzrignore
+.hg/
+.hgignore
+.svn/
+# Common backup files
+*.swp
+*.bak
+*.tmp
+*.orig
+*~
+# Various IDEs
+.project
+.idea/
+*.tmproj
+.vscode/

helm-charts/common/vllm/Chart.yaml

@@ -0,0 +1,10 @@
+# Copyright (C) 2024 Intel Corporation
+# SPDX-License-Identifier: Apache-2.0
+
+apiVersion: v2
+name: vllm
+description: The Helm chart for vLLM Inference Server
+type: application
+version: 1.0.0
+# The vLLM version
+appVersion: "0.5"

helm-charts/common/vllm/README.md

@@ -0,0 +1,53 @@
+# vllm
+
+Helm chart for deploying vLLM Inference service.
+
+Refer to [Deploy with Helm Charts](../README.md) for global guides.
+
+## Installing the Chart
+
+To install the chart, run the following:
+
+Note that you cannot use vllm as the service release name due to [environment variables conflict](https://docs.vllm.ai/en/stable/serving/env_vars.html#environment-variables).
+
+```console
+cd GenAIInfra/helm-charts/common
+export MODELDIR=/mnt/opea-models
+export MODELNAME="Intel/neural-chat-7b-v3-3"
+export HFTOKEN="insert-your-huggingface-token-here"
+helm install myvllm vllm --set global.modelUseHostPath=${MODELDIR} --set LLM_MODEL_ID=${MODELNAME} --set global.HUGGINGFACEHUB_API_TOKEN=${HFTOKEN}
+# To deploy on Gaudi enabled kubernetes cluster
+# helm install myvllm vllm --set global.modelUseHostPath=${MODELDIR} --set LLM_MODEL_ID=${MODELNAME} --set global.HUGGINGFACEHUB_API_TOKEN=${HFTOKEN} --values gaudi-values.yaml
+```
+
+By default, the vllm service will downloading the "Intel/neural-chat-7b-v3-3".
+
+If you already cached the model locally, you can pass it to container like this example:
+
+MODELDIR=/mnt/opea-models
+
+MODELNAME="facebook/opt-125m"
+
+## Verify
+
+To verify the installation, run the command `kubectl get pod` to make sure all pods are runinng.
+
+Then run the command `kubectl port-forward svc/myvllm 2080:80` to expose the vllm service for access.
+
+Open another terminal and run the following command to verify the service if working:
+
+```console
+curl http://localhost:2080/v1/completions \
+  -H "Content-Type: application/json" \
+  -d '{"model": "Intel/neural-chat-7b-v3-3", "prompt": "What is Deep Learning?", "max_tokens": 32, "temperature": 0}'
+```
+
+## Values
+
+| Key                             | Type   | Default                              | Description                                                                                                                                                                                                            |
+| ------------------------------- | ------ | ------------------------------------ | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| LLM_MODEL_ID                    | string | `"Intel/neural-chat-7b-v3-3"`        | Models id from https://huggingface.co/, or predownloaded model directory                                                                                                                                               |
+| global.HUGGINGFACEHUB_API_TOKEN | string | `insert-your-huggingface-token-here` | Hugging Face API token                                                                                                                                                                                                 |
+| global.modelUseHostPath         | string | `""`                                 | Cached models directory, vllm will not download if the model is cached here. The host path "modelUseHostPath" will be mounted to container as /data directory. Set this to null/empty will force it to download model. |
+| image.repository                | string | `"opea/vllm"`                        |                                                                                                                                                                                                                        |
+| image.tag                       | string | `"latest"`                           |                                                                                                                                                                                                                        |

helm-charts/common/vllm/gaudi-values.yaml

@@ -0,0 +1,19 @@
+# Copyright (C) 2024 Intel Corporation
+# SPDX-License-Identifier: Apache-2.0
+
+# Default values for vllm.
+# This is a YAML-formatted file.
+# Declare variables to be passed into your templates.
+
+image:
+  repository: opea/llm-vllm-hpu
+  tag: "latest"
+
+VLLM_CPU_KVCACHE_SPACE: "40"
+
+# Workaround for current HPU image with start command /bin/bash
+# extraCmdArgs: ["--enforce-eager","--tensor-parallel-size","1","--block-size","128","--max-num-seqs","256","--max-seq_len-to-capture","2048"]
+extraCmdArgs: ["/bin/bash","-c","python3 -m vllm.entrypoints.openai.api_server --enforce-eager --model Intel/neural-chat-7b-v3-3 --tensor-parallel-size 1 --host 0.0.0.0 --port 2080 --download-dir /data --block-size 128 --max-num-seqs 256 --max-seq_len-to-capture 2048"]
+resources:
+  limits:
+    habana.ai/gaudi: 1

helm-charts/common/vllm/templates/_helpers.tpl

@@ -0,0 +1,62 @@
+{{/*
+Expand the name of the chart.
+*/}}
+{{- define "vllm.name" -}}
+{{- default .Chart.Name .Values.nameOverride | trunc 63 | trimSuffix "-" }}
+{{- end }}
+
+{{/*
+Create a default fully qualified app name.
+We truncate at 63 chars because some Kubernetes name fields are limited to this (by the DNS naming spec).
+If release name contains chart name it will be used as a full name.
+*/}}
+{{- define "vllm.fullname" -}}
+{{- if .Values.fullnameOverride }}
+{{- .Values.fullnameOverride | trunc 63 | trimSuffix "-" }}
+{{- else }}
+{{- $name := default .Chart.Name .Values.nameOverride }}
+{{- if contains $name .Release.Name }}
+{{- .Release.Name | trunc 63 | trimSuffix "-" }}
+{{- else }}
+{{- printf "%s-%s" .Release.Name $name | trunc 63 | trimSuffix "-" }}
+{{- end }}
+{{- end }}
+{{- end }}
+
+{{/*
+Create chart name and version as used by the chart label.
+*/}}
+{{- define "vllm.chart" -}}
+{{- printf "%s-%s" .Chart.Name .Chart.Version | replace "+" "_" | trunc 63 | trimSuffix "-" }}
+{{- end }}
+
+{{/*
+Common labels
+*/}}
+{{- define "vllm.labels" -}}
+helm.sh/chart: {{ include "vllm.chart" . }}
+{{ include "vllm.selectorLabels" . }}
+{{- if .Chart.AppVersion }}
+app.kubernetes.io/version: {{ .Chart.AppVersion | quote }}
+{{- end }}
+app.kubernetes.io/managed-by: {{ .Release.Service }}
+{{- end }}
+
+{{/*
+Selector labels
+*/}}
+{{- define "vllm.selectorLabels" -}}
+app.kubernetes.io/name: {{ include "vllm.name" . }}
+app.kubernetes.io/instance: {{ .Release.Name }}
+{{- end }}
+
+{{/*
+Create the name of the service account to use
+*/}}
+{{- define "vllm.serviceAccountName" -}}
+{{- if .Values.serviceAccount.create }}
+{{- default (include "vllm.fullname" .) .Values.serviceAccount.name }}
+{{- else }}
+{{- default "default" .Values.serviceAccount.name }}
+{{- end }}
+{{- end }}

helm-charts/common/vllm/templates/configmap.yaml

@@ -0,0 +1,25 @@
+# Copyright (C) 2024 Intel Corporation
+# SPDX-License-Identifier: Apache-2.0
+
+apiVersion: v1
+kind: ConfigMap
+metadata:
+  name: {{ include "vllm.fullname" . }}-config
+  labels:
+    {{- include "vllm.labels" . | nindent 4 }}
+data:
+  HF_TOKEN: {{ .Values.global.HUGGINGFACEHUB_API_TOKEN | quote}}
+  {{- if .Values.global.HF_ENDPOINT }}
+  HF_ENDPOINT: {{ .Values.global.HF_ENDPOINT | quote}}
+  {{- end }}
+  http_proxy: {{ .Values.global.http_proxy | quote }}
+  https_proxy: {{ .Values.global.https_proxy | quote }}
+  no_proxy: {{ .Values.global.no_proxy | quote }}
+  HABANA_LOGS: "/tmp/habana_logs"
+  NUMBA_CACHE_DIR: "/tmp"
+  HF_HOME: "/tmp/.cache/huggingface"
+  # https://github.com/outlines-dev/outlines/blob/main/outlines/caching.py#L14-L29
+  OUTLINES_CACHE_DIR: "/tmp/.cache/outlines"
+  {{- if .Values.VLLM_CPU_KVCACHE_SPACE }}
+  VLLM_CPU_KVCACHE_SPACE: {{ .Values.VLLM_CPU_KVCACHE_SPACE | quote}}
+  {{- end }}

helm-charts/common/vllm/templates/deployment.yaml

@@ -0,0 +1,116 @@
+# Copyright (C) 2024 Intel Corporation
+# SPDX-License-Identifier: Apache-2.0
+
+apiVersion: apps/v1
+kind: Deployment
+metadata:
+  name: {{ include "vllm.fullname" . }}
+  labels:
+    {{- include "vllm.labels" . | nindent 4 }}
+spec:
+  replicas: {{ .Values.replicaCount }}
+  selector:
+    matchLabels:
+      {{- include "vllm.selectorLabels" . | nindent 6 }}
+  template:
+    metadata:
+      {{- with .Values.podAnnotations }}
+      annotations:
+        {{- toYaml . | nindent 8 }}
+      {{- end }}
+      labels:
+        {{- include "vllm.selectorLabels" . | nindent 8 }}
+    spec:
+      {{- with .Values.imagePullSecrets }}
+      imagePullSecrets:
+        {{- toYaml . | nindent 8 }}
+      {{- end }}
+      securityContext:
+        {{- toYaml .Values.podSecurityContext | nindent 8 }}
+      containers:
+        - name: {{ .Chart.Name }}
+          envFrom:
+            - configMapRef:
+                name: {{ include "vllm.fullname" . }}-config
+            {{- if .Values.global.extraEnvConfig }}
+            - configMapRef:
+                name: {{ .Values.global.extraEnvConfig }}
+                optional: true
+            {{- end }}
+          securityContext:
+            {{- if .Values.global.modelUseHostPath }}
+            {}
+            {{- else }}
+            {{- toYaml .Values.securityContext | nindent 12 }}
+            {{- end }}
+          image: "{{ .Values.image.repository }}:{{ .Values.image.tag | default .Chart.AppVersion }}"
+          imagePullPolicy: {{ .Values.image.pullPolicy }}
+          args:
+          {{- if .Values.extraCmdArgs }}
+            {{- range .Values.extraCmdArgs }}
+            - {{ . | quote }}
+            {{- end }}
+          {{- end }}
+            - "--model"
+            - {{ .Values.LLM_MODEL_ID | quote }}
+            - "--host"
+            - "0.0.0.0"
+            - "--port"
+            - {{ .Values.port | quote }}
+            - "--download-dir"
+            - "/data"
+          volumeMounts:
+            - mountPath: /data
+              name: model-volume
+            - mountPath: /dev/shm
+              name: shm
+            - mountPath: /tmp
+              name: tmp
+          ports:
+            - name: http
+              containerPort: {{ .Values.port }}
+              protocol: TCP
+          {{- if .Values.livenessProbe }}
+          livenessProbe:
+            {{- toYaml .Values.livenessProbe | nindent 12 }}
+          {{- end }}
+          {{- if .Values.readinessProbe }}
+          readinessProbe:
+            {{- toYaml .Values.readinessProbe | nindent 12 }}
+          {{- end }}
+          {{- if .Values.startupProbe }}
+          startupProbe:
+            {{- toYaml .Values.startupProbe | nindent 12 }}
+          {{- end }}
+          resources:
+            {{- toYaml .Values.resources | nindent 12 }}
+      volumes:
+        - name: model-volume
+          {{- if .Values.global.modelUsePVC }}
+          persistentVolumeClaim:
+            claimName: {{ .Values.global.modelUsePVC }}
+          {{- else if .Values.global.modelUseHostPath }}
+          hostPath:
+            path: {{ .Values.global.modelUseHostPath }}
+            type: Directory
+          {{- else }}
+          emptyDir: {}
+          {{- end }}
+        - name: shm
+          emptyDir:
+            medium: Memory
+            sizeLimit: {{ .Values.shmSize }}
+        - name: tmp
+          emptyDir: {}
+      {{- with .Values.nodeSelector }}
+      nodeSelector:
+        {{- toYaml . | nindent 8 }}
+      {{- end }}
+      {{- with .Values.affinity }}
+      affinity:
+        {{- toYaml . | nindent 8 }}
+      {{- end }}
+      {{- with .Values.tolerations }}
+      tolerations:
+        {{- toYaml . | nindent 8 }}
+      {{- end }}

helm-charts/common/vllm/templates/service.yaml

@@ -0,0 +1,18 @@
+# Copyright (C) 2024 Intel Corporation
+# SPDX-License-Identifier: Apache-2.0
+
+apiVersion: v1
+kind: Service
+metadata:
+  name: {{ include "vllm.fullname" . }}
+  labels:
+    {{- include "vllm.labels" . | nindent 4 }}
+spec:
+  type: {{ .Values.service.type }}
+  ports:
+    - port: 80
+      targetPort: {{ .Values.port }}
+      protocol: TCP
+      name: vllm
+  selector:
+    {{- include "vllm.selectorLabels" . | nindent 4 }}

helm-charts/common/vllm/templates/tests/test-pod.yaml

@@ -0,0 +1,29 @@
+# Copyright (C) 2024 Intel Corporation
+# SPDX-License-Identifier: Apache-2.0
+
+apiVersion: v1
+kind: Pod
+metadata:
+  name: "{{ include "vllm.fullname" . }}-testpod"
+  labels:
+    {{- include "vllm.labels" . | nindent 4 }}
+  annotations:
+    "helm.sh/hook": test
+    #"helm.sh/hook-delete-policy": "hook-succeeded, hook-failure"
+spec:
+  containers:
+    - name: curl
+      image: python:3.10.14
+      command: ['bash', '-c']
+      args:
+        - |
+          max_retry=20;
+          for ((i=1; i<=max_retry; i++)); do
+            curl http://{{ include "vllm.fullname" . }}/v1/completions \
+            -H "Content-Type: application/json" \
+            -d '{"model": {{ .Values.LLM_MODEL_ID | quote }},"prompt": "What is Deep Learning?","max_tokens": 32,"temperature": 0}' && break;
+            curlcode=$?
+            if [[ $curlcode -eq 7 ]]; then sleep 10; else echo "curl failed with code $curlcode"; exit 1; fi;
+          done;
+          if [ $i -gt $max_retry ]; then echo "test failed with maximum retry"; exit 1; fi
+  restartPolicy: Never