Skip to content

Commit

Permalink
Initial commit for CodeGen on OpenShift
Browse files Browse the repository at this point in the history
Signed-off-by: Paulina Olszewska <[email protected]>
  • Loading branch information
polszewska committed Oct 29, 2024
1 parent 593458c commit 5dec0dd
Show file tree
Hide file tree
Showing 73 changed files with 2,262 additions and 0 deletions.
12 changes: 12 additions & 0 deletions helm-charts/codegen-openshift-rhoai/Chart.yaml
Original file line number Diff line number Diff line change
@@ -0,0 +1,12 @@
apiVersion: v2
name: codegen
description: A Helm chart for deploying codegen on Red Hat OpenShift with Red Hat OpenShift AI
dependencies:
- name: llm-uservice
version: 1.0.0
repository: "file://llm-uservice"
- name: react-ui
version: 0.1.0
repository: "file://react-ui"
type: application
version: 1.0.0
111 changes: 111 additions & 0 deletions helm-charts/codegen-openshift-rhoai/README.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,111 @@
# CodeGen

Helm chart for deploying CodeGen service on Red Hat OpenShift with Red Hat OpenShift AI.

Serving runtime template in this example uses model *ise-uiuc/Magicoder-S-DS-6.7B* for Xeon and *meta-llama/CodeLlama-7b-hf* for Gaudi.

## Prerequisites

1. **Red Hat OpenShift Cluster** with dynamic *StorageClass* to provision *PersistentVolumes* e.g. **OpenShift Data Foundation**) and installed Operators: **Red Hat - Authorino (Technical Preview)**, **Red Hat OpenShift Service Mesh**, **Red Hat OpenShift Serverless** and **Red Hat Openshift AI**.
2. Image registry to push there docker images (https://docs.openshift.com/container-platform/4.16/registry/securing-exposing-registry.html).
3. Access to S3-compatible object storage bucket (e.g. **OpenShift Data Foundation**, **AWS S3**) and values of access and secret access keys and S3 endpoint (https://docs.redhat.com/en/documentation/red_hat_openshift_data_foundation/4.16/html/managing_hybrid_and_multicloud_resources/accessing-the-multicloud-object-gateway-with-your-applications_rhodf#accessing-the-multicloud-object-gateway-with-your-applications_rhodf).
4. Account on https://huggingface.co/, access to model *ise-uiuc/Magicoder-S-DS-6.7B* (for Xeon) or *meta-llama/CodeLlama-7b-hf* (for Gaudi) and token with Read permissions.

## Deploy model in Red Hat Openshift AI

1. Login to OpenShift CLI and run following commands to create new serving runtime and *hf-token* secret.
```
cd GenAIInfra/helm-charts/codegen-openshift-rhoai/
export HFTOKEN="insert-your-huggingface-token-here"
On Xeon:
helm install servingruntime tgi --set global.huggingfacehubApiToken=${HFTOKEN}
On Gaudi:
helm install servingruntime tgi --set global.huggingfacehubApiToken=${HFTOKEN} --values tgi/gaudi-values.yaml
```

Verify if template has been created with ```oc get template -n redhat-ods-applications``` command.

2. Find the route for **Red Hat OpenShift AI** dashboard with below command and open it in the browser:
```
oc get routes -A | grep rhods-dashboard
```
3. Go to **Data Science Project** and clik **Create data science project**. Fill the **Name** and click **Create**.
4. Go to **Workbenches** tab and clik **Create workbench**. Fill the **Name**, under **Notebook image** choose *Standard Data Science*, under **Cluster storage** choose *Create new persistent storage* and change **Persistent storage size** to 40 GB. Click **Create workbench**.
5. Open newly created Jupiter notebook and run following commands to download the model and upload it on s3:
```
%env S3_ENDPOINT=<S3_RGW_ROUTE>
%env S3_ACCESS_KEY=<AWS_ACCESS_KEY_ID>
%env S3_SECRET_KEY=<AWS_SECRET_ACCESS_KEY>
%env HF_TOKEN=<PASTE_HUGGINGFACE_TOKEN>
```
```
!pip install huggingface-hub
```
```
import os
import boto3
import botocore
import glob
from huggingface_hub import snapshot_download
bucket_name = 'first.bucket'
s3_endpoint = os.environ.get('S3_ENDPOINT')
s3_accesskey = os.environ.get('S3_ACCESS_KEY')
s3_secretkey = os.environ.get('S3_SECRET_KEY')
path = 'models'
hf_token = os.environ.get('HF_TOKEN')
session = boto3.session.Session()
s3_resource = session.resource('s3',
endpoint_url=s3_endpoint,
verify=False,
aws_access_key_id=s3_accesskey,
aws_secret_access_key=s3_secretkey)
bucket = s3_resource.Bucket(bucket_name)
```
For Xeon download *ise-uiuc/Magicoder-S-DS-6.7B*:
```
snapshot_download("ise-uiuc/Magicoder-S-DS-6.7B", cache_dir=f'./models', token=hf_token)
```
For Gaudi download *meta-llama/CodeLlama-7b-hf*:
```
snapshot_download("meta-llama/CodeLlama-7b-hf", cache_dir=f'./models', token=hf_token)
```
Upload the downloaded model to S3:
```
files = (file for file in glob.glob(f'{path}/**/*', recursive=True) if os.path.isfile(file) and "snapshots" in file)
for filename in files:
s3_name = filename.replace(path, '')
print(f'Uploading: {filename} to {path}{s3_name}')
bucket.upload_file(filename, f'{path}{s3_name}')
```

6. Go to your project in **Red Hat OpenShift AI** dashboard, then "Models" tab and click **Deploy model** under *Single-model serving platform*. Fill the **Name**, choose newly created **Serving runtime**: *Text Generation Inference Magicoder-S-DS-6.7B on CPU* (for Xeon) or *Text Generation Inference CodeLlama-7b-hf on Gaudi* (for Gaudi), **Model framework**: *llm* and change **Model server size** to *Custom*: 16 CPUs and 64 Gi memory. For deployment with Gaudi select proper **Accelerator**. Click the checkbox to create external route in **Model route** section and uncheck the **Token authentication**. Under **Model location** choose *New data connection* and fill all required fields for s3 access, **Bucket** *first.bucket* and **Path**: *models*. Click **Deploy**. It takes about 10 minutes to get *Loaded* status.\
If it's not going to *Loaded* status and revision changed status to "ProgressDeadlineExceeded" (``oc get revision``), scale model deployment to 0 and than to 1 with command ``oc scale deployment.apps/<model_deployment_name> --replicas=1`` and wait about 10 minutes for deployment.

## Install the Chart

To install the chart, back to OpenShift CLI, go to your project and run the following:

```console
cd GenAIInfra/helm-charts/

export NAMESPACE="insert-your-namespace-here"
export CLUSTERDOMAIN="$(oc get Ingress.config.openshift.io/cluster -o jsonpath='{.spec.domain}' | sed 's/^apps.//')"
export MODELNAME="insert-name-of-deployed-model-here" (it refers to the *Name* from step 6 in **Deploy model in Red Hat Openshift AI**)
export PROJECT="insert-project-name-where-model-is-deployed"

sed -i "s/insert-your-namespace-here/${NAMESPACE}/g" codegen-openshift-rhoai/llm-uservice/values.yaml

./update_dependency.sh
helm dependency update codegen-openshift-rhoai

helm install codegen codegen-openshift-rhoai --set image.repository=image-registry.openshift-image-registry.svc:5000/${NAMESPACE}/codegen --set llm-uservice.image.repository=image-registry.openshift-image-registry.svc:5000/${NAMESPACE}/llm-tgi --set react-ui.image.repository=image-registry.openshift-image-registry.svc:5000/${NAMESPACE}/react-ui --set global.clusterDomain=${CLUSTERDOMAIN} --set global.huggingfacehubApiToken=${HFTOKEN} --set llm-uservice.servingRuntime.name=${MODELNAME} --set llm-uservice.servingRuntime.namespace=${PROJECT}
```

## Verify

To verify the installation, run the command `oc get pods` to make sure all pods are running. Wait about 5 minutes for building images. When 4 pods achieve *Completed* status, the rest with services should go to *Running*.

## Launch the UI
To access the frontend, find the route for *react-ui* with command `oc get routes` and open it in the browser.
5 changes: 5 additions & 0 deletions helm-charts/codegen-openshift-rhoai/llm-uservice/Chart.yaml
Original file line number Diff line number Diff line change
@@ -0,0 +1,5 @@
apiVersion: v2
name: llm-uservice
description: A Helm chart for deploying llm-uservice on Red Hat OpenShift with Red Hat OpenShift AI
type: application
version: 1.0.0
Original file line number Diff line number Diff line change
@@ -0,0 +1,51 @@
{{/*
Expand the name of the chart.
*/}}
{{- define "llm-uservice.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 "llm-uservice.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 "llm-uservice.chart" -}}
{{- printf "%s-%s" .Chart.Name .Chart.Version | replace "+" "_" | trunc 63 | trimSuffix "-" }}
{{- end }}

{{/*
Common labels
*/}}
{{- define "llm-uservice.labels" -}}
helm.sh/chart: {{ include "llm-uservice.chart" . }}
{{ include "llm-uservice.selectorLabels" . }}
{{- if .Chart.AppVersion }}
app.kubernetes.io/version: {{ .Chart.AppVersion | quote }}
{{- end }}
app.kubernetes.io/managed-by: {{ .Release.Service }}
{{- end }}

{{/*
Selector labels
*/}}
{{- define "llm-uservice.selectorLabels" -}}
app.kubernetes.io/name: {{ include "llm-uservice.name" . }}
app.kubernetes.io/instance: {{ .Release.Name }}
{{- end }}
Original file line number Diff line number Diff line change
@@ -0,0 +1,27 @@
kind: BuildConfig
apiVersion: build.openshift.io/v1
metadata:
name: {{ include "llm-uservice.fullname" . }}
namespace: {{ .Release.Namespace }}
spec:
output:
to:
kind: "ImageStreamTag"
name: "llm-tgi:latest"
failedBuildsHistoryLimit: 5
successfulBuildsHistoryLimit: 5
nodeSelector: null
postCommit: {}
resources: {}
runPolicy: SerialLatestOnly
source:
git:
ref: {{ .Values.source.gitRef }}
uri: {{ .Values.source.gitUri }}
type: {{ .Values.source.type }}
strategy:
type: {{ .Values.strategy.type }}
dockerStrategy:
dockerfilePath: {{ .Values.strategy.dockerfilePath }}
triggers:
- type: ConfigChange
Original file line number Diff line number Diff line change
@@ -0,0 +1,14 @@
apiVersion: v1
kind: ConfigMap
metadata:
name: create-rhoai-istio-rootca-cert-secret
data:
create-rhoai-istio-rootca-cert-secret.sh: |
#!/bin/bash
EXISTS=$(oc get secret --ignore-not-found rhoai-ca-bundle)
if [[ -z "${EXISTS}" ]]; then
oc create secret generic -n {{ .Release.Namespace }} rhoai-ca-bundle --from-literal=tls.crt="$(oc extract secret/knative-serving-cert -n istio-system --to=- --keys=tls.crt)"
else
echo "oc get secret --ignore-not-found rhoai-ca-bundle returned non-empty string, not creating a secret"
fi
Original file line number Diff line number Diff line change
@@ -0,0 +1,63 @@
apiVersion: apps/v1
kind: Deployment
metadata:
name: {{ include "llm-uservice.fullname" . }}
labels:
{{- include "llm-uservice.labels" . | nindent 4 }}
annotations:
image.openshift.io/triggers: '[{"from":{"kind":"ImageStreamTag","name":"llm-tgi:latest"},"fieldPath":"spec.template.spec.containers[?(@.name==\"codegen\")].image"}]'
spec:
replicas: 1
selector:
matchLabels:
{{- include "llm-uservice.selectorLabels" . | nindent 6 }}
template:
metadata:
labels:
{{- include "llm-uservice.selectorLabels" . | nindent 8 }}
spec:
securityContext: {}
containers:
- name: {{ .Release.Name }}
command:
- /bin/bash
- -c
- |
cp /usr/lib/ssl/cert.pem /tmp/bundle.crt && \
cat /rhoai-ca/tls.crt | tee -a '/tmp/bundle.crt' && \
bash ./entrypoint.sh
env:
- name: TGI_LLM_ENDPOINT
value: "https://{{ .Values.servingRuntime.name }}-{{ .Values.servingRuntime.namespace }}.apps.{{ .Values.global.clusterDomain }}"
- name: HUGGINGFACEHUB_API_TOKEN
valueFrom:
secretKeyRef:
key: HUGGING_FACE_HUB_TOKEN
name: hf-token
- name: PYTHONPATH
value: {{ .Values.PYTHONPATH | quote }}
- name: HOME
value: {{ .Values.HOME | quote }}
- name: SSL_CERT_FILE
value: /tmp/bundle.crt
securityContext: {}
image: "{{ .Values.image.repository }}:{{ .Values.image.tag }}"
imagePullPolicy: {{ .Values.image.pullPolicy }}
ports:
- name: llm-uservice
containerPort: 9000
protocol: TCP
volumeMounts:
- mountPath: /tmp/home
name: local-dir
- mountPath: /rhoai-ca
name: odh-ca-bundle
resources: {}
volumes:
- emptyDir:
sizeLimit: 5Gi
name: local-dir
- name: odh-ca-bundle
secret:
defaultMode: 420
secretName: rhoai-ca-bundle
Original file line number Diff line number Diff line change
@@ -0,0 +1,9 @@
---
apiVersion: image.openshift.io/v1
kind: ImageStream
metadata:
name: llm-tgi
namespace: {{ .Release.Namespace }}
spec:
lookupPolicy:
local: true
Original file line number Diff line number Diff line change
@@ -0,0 +1,31 @@
apiVersion: batch/v1
kind: Job
metadata:
name: create-rhoai-istio-rootca-cert-secret
spec:
template:
spec:
containers:
- image: {{ .Values.job.image }}
command:
- /bin/bash
- -c
- |
oc wait --for=condition=ReconcileComplete=True dsc/rhods-datasciencecluster --timeout=-1s
oc wait --for condition=Ready=True knativeserving -n knative-serving knative-serving --timeout=-1s
'/tmp/create-rhoai-istio-rootca-cert-secret.sh'
name: create-rhoai-istio-rootca-cert-secret
volumeMounts:
- mountPath: /tmp/create-rhoai-istio-rootca-cert-secret.sh
name: create-rhoai-istio-rootca-cert-secret
subPath: create-rhoai-istio-rootca-cert-secret.sh
volumes:
- name: create-rhoai-istio-rootca-cert-secret
configMap:
name: create-rhoai-istio-rootca-cert-secret
defaultMode: 0755
dnsPolicy: ClusterFirst
restartPolicy: Never
serviceAccount: {{ .Values.serviceAccountName }}
serviceAccountName: {{ .Values.serviceAccountName }}
terminationGracePeriodSeconds: 400
Original file line number Diff line number Diff line change
@@ -0,0 +1,12 @@
apiVersion: rbac.authorization.k8s.io/v1
kind: ClusterRoleBinding
metadata:
name: kubeadmin-rhoai-cluster-admin-rolebinding
roleRef:
apiGroup: rbac.authorization.k8s.io
kind: ClusterRole
name: cluster-admin
subjects:
- kind: User
apiGroup: rbac.authorization.k8s.io
name: 'kube:admin'
Original file line number Diff line number Diff line change
@@ -0,0 +1,28 @@
{{- range $key, $value := .Values.rbac.roles }}
{{- if $value.createRole }}
---
apiVersion: rbac.authorization.k8s.io/v1
{{- if $value.scope.cluster }}
kind: ClusterRole
metadata:
name: {{ .name }}
{{- else }}
kind: Role
metadata:
name: {{ $value.name }}
{{- end }}
rules:
- apiGroups:
{{- range $value.apiGroups }}
- {{ . }}
{{- end }}
resources:
{{- range $value.resources }}
- {{ . }}
{{- end }}
verbs:
{{- range $value.verbs }}
- {{ . }}
{{- end }}
{{- end }}
{{- end }}
Original file line number Diff line number Diff line change
@@ -0,0 +1,25 @@
{{- range $key, $value := .Values.rbac.roleBindings }}
{{- if $value.createBinding }}
apiVersion: rbac.authorization.k8s.io/v1
{{- if $value.scope.cluster }}
kind: ClusterRoleBinding
{{- else }}
kind: RoleBinding
{{- end }}
metadata:
name: {{ .name }}
{{- if eq $value.scope.cluster false }}
namespace: {{ $value.scope.namespace }}
{{- end }}
subjects:
- kind: {{ $value.subjects.kind | default "ServiceAccount" }}
name: {{ $value.subjects.name }}
namespace: {{ $value.subjects.namespace }}
apiGroup: ""
roleRef:
kind: {{ $value.roleRef.kind }}
name: {{ $value.roleRef.name }}
apiGroup: rbac.authorization.k8s.io
---
{{- end }}
{{- end }}
Loading

0 comments on commit 5dec0dd

Please sign in to comment.