Installation

Requirements

Python 3.8 or higher
Node.js 16 or higher
Docker (optional)

You have the option to install rlhfblender either locally or via Docker. The Docker installation is recommended for users who do not have Python or Node.js installed on their system.

Clone the repository

git clone https://github.com/ymetz/rlhfblender.git
cd rlhfblender
git submodule update --init rlhfblender-ui

to get both the main repository,user interface. If you want to download both the repository and demo models, you can also run git clone --recurse-submodules https://github.com/ymetz/rlhfblender.git.

Local Installation

cd rlhfblender
pip install -r requirements.txt
python rlhfblender/app.py

and

cd rlhfblender-ui
npm install
npm run start

Docker Installation

docker-compose up

Usage

After starting the aplication: The user interface is available at http://localhost:3000. The API is available at http://localhost:8080/docs.

Kubernetes Deployment

The following is an example of a Kubernetes deployment. The deployment contains two containers, one for the API and one for the user interface.

Backend/API

app:
image:
    repository: "<YOUR_REPOSTIRY>/rlhfblender-backend"
    tag: latest

replicaCount: 1

regcred: regcred-rlworkbench
port: 8080

livenessProbe: "null"

readinessProbe: "null"

startupProbe: "null"

requests:
    cpu: 100m
    memory: 500Mi
    # Optional: If you want to use a GPU
    gpu: 1
limits:
    cpu: 4000m
    memory: 12Gi

gpu:
    devices: 0,...

extraEnv:
    BACKEND_PORT: "{{ .Values.app.port }}"

ingress:
    enabled: false

Frontend/UI

app:
image:
    repository: '${CI_REGISTRY_IMAGE}/frontend'
    tag: '$VERSION'
replicaCount: $REPLICA_COUNT
regcred: regcred-rlworkbench
port: 3000

requests:
    cpu: 100m
    memory: 250Mi
limits:
    cpu: 1000m
    memory: 4Gi

livenessProbe: |
    httpGet:
    path: "/"
    port: {{ .Values.app.port }}
    scheme: HTTP
    initialDelaySeconds: 60
    timeoutSeconds: 10
    periodSeconds: 30
    failureThreshold: 3
    successThreshold: 1

readinessProbe: |
    httpGet:
    path: "/"
    port: {{ .Values.app.port }}
    scheme: HTTP
    initialDelaySeconds: 60
    timeoutSeconds: 10
    periodSeconds: 30
    failureThreshold: 3
    successThreshold: 1

extraEnv:
    # Choose hostname
    BACKEND_HOST: 'rlhfblender-backend'
    BACKEND_PORT: '8080'
    # Choose URL
    HOST: 'rlhfblender.example.com'

ingress:
    enabled: true
    url: 'rlhfblender.example.com'
    extraAnnotations: |
    nginx.ingress.kubernetes.io/proxy-body-size: 8m