mirror of
https://github.com/suitenumerique/docs.git
synced 2026-10-01 05:55:16 +02:00
lebaudantoine
8b60bc57e6
simplify LLM's job. Do not request Json output with a single key. Instead, make sure LLM don't output any extra information. By simplifying LLM's job, we're making sure its output can be parsed. I did a quick test with the Translate prompt. Adding instructions to output only translated text seems enough after a bunch of tests. I did a small prompt engineering, using ChatGPT and Claude to generate a proper system prompt … it works quite okay BUT there is room for improvement for sure. I'ven't searched yet OS prompts we could find in a prompt library. Perfect translation job seems to be a difficult job for a 8B model. Please note I haven't updated yet the other prompts, let's discuss it before. I ran my experiment with our internal LLM which is optimized for throughput, and not latency (there is a trade-off). I'll try fine tune few of its parameters to see if I can reduce its latency. For 880 tokens (based on chatgpt tokens counter online). It takes roughly 17s, vs ~40s for Albert CNRS 70B. For 180 tokens it takes roughly 3s. Without a proper UX (eg. a nicer loading animation, streaming tokens) it feels a decade. However, asking Chatgpt the same job take the same amount, from submitting the request to the last token being generated.
141 lines
4.3 KiB
Go Template
141 lines
4.3 KiB
Go Template
image:
|
|
repository: localhost:5001/impress-backend
|
|
pullPolicy: Always
|
|
tag: "latest"
|
|
|
|
backend:
|
|
replicas: 1
|
|
envVars:
|
|
COLLABORATION_API_URL: https://impress.127.0.0.1.nip.io/collaboration/api/
|
|
COLLABORATION_SERVER_SECRET: my-secret
|
|
DJANGO_CSRF_TRUSTED_ORIGINS: https://impress.127.0.0.1.nip.io
|
|
DJANGO_CONFIGURATION: Feature
|
|
DJANGO_ALLOWED_HOSTS: impress.127.0.0.1.nip.io
|
|
DJANGO_SECRET_KEY: {{ .Values.djangoSecretKey }}
|
|
DJANGO_SETTINGS_MODULE: impress.settings
|
|
DJANGO_SUPERUSER_PASSWORD: admin
|
|
DJANGO_EMAIL_HOST: "mailcatcher"
|
|
DJANGO_EMAIL_PORT: 1025
|
|
DJANGO_EMAIL_USE_SSL: False
|
|
LOGGING_LEVEL_HANDLERS_CONSOLE: ERROR
|
|
LOGGING_LEVEL_LOGGERS_ROOT: INFO
|
|
LOGGING_LEVEL_LOGGERS_APP: INFO
|
|
OIDC_OP_JWKS_ENDPOINT: https://fca.integ01.dev-agentconnect.fr/api/v2/jwks
|
|
OIDC_OP_AUTHORIZATION_ENDPOINT: https://fca.integ01.dev-agentconnect.fr/api/v2/authorize
|
|
OIDC_OP_TOKEN_ENDPOINT: https://fca.integ01.dev-agentconnect.fr/api/v2/token
|
|
OIDC_OP_USER_ENDPOINT: https://fca.integ01.dev-agentconnect.fr/api/v2/userinfo
|
|
OIDC_OP_LOGOUT_ENDPOINT: https://fca.integ01.dev-agentconnect.fr/api/v2/session/end
|
|
OIDC_RP_CLIENT_ID: {{ .Values.oidc.clientId }}
|
|
OIDC_RP_CLIENT_SECRET: {{ .Values.oidc.clientSecret }}
|
|
OIDC_RP_SIGN_ALGO: RS256
|
|
OIDC_RP_SCOPES: "openid email given_name usual_name"
|
|
USER_OIDC_FIELD_TO_SHORTNAME: "given_name"
|
|
USER_OIDC_FIELDS_TO_FULLNAME: "given_name,usual_name"
|
|
OIDC_REDIRECT_ALLOWED_HOSTS: https://impress.127.0.0.1.nip.io
|
|
OIDC_AUTH_REQUEST_EXTRA_PARAMS: "{'acr_values': 'eidas1'}"
|
|
LOGIN_REDIRECT_URL: https://impress.127.0.0.1.nip.io
|
|
LOGIN_REDIRECT_URL_FAILURE: https://impress.127.0.0.1.nip.io
|
|
LOGOUT_REDIRECT_URL: https://impress.127.0.0.1.nip.io
|
|
DB_HOST: postgres-postgresql
|
|
DB_NAME: impress
|
|
DB_USER: dinum
|
|
DB_PASSWORD: pass
|
|
DB_PORT: 5432
|
|
POSTGRES_DB: impress
|
|
POSTGRES_USER: dinum
|
|
POSTGRES_PASSWORD: pass
|
|
REDIS_URL: redis://default:pass@redis-master:6379/1
|
|
AWS_S3_ENDPOINT_URL: http://minio.impress.svc.cluster.local:9000
|
|
AWS_S3_ACCESS_KEY_ID: impress
|
|
AWS_S3_SECRET_ACCESS_KEY: password
|
|
AWS_STORAGE_BUCKET_NAME: impress-media-storage
|
|
STORAGES_STATICFILES_BACKEND: django.contrib.staticfiles.storage.StaticFilesStorage
|
|
AI_API_KEY: **ask antoine**
|
|
AI_BASE_URL: https://albertine.beta.numerique.gouv.fr/v1/
|
|
AI_MODEL: meta-llama/Llama-3.1-8B-Instruct
|
|
|
|
migrate:
|
|
command:
|
|
- "/bin/sh"
|
|
- "-c"
|
|
- |
|
|
python manage.py migrate --no-input &&
|
|
python manage.py create_demo --force
|
|
restartPolicy: Never
|
|
|
|
command:
|
|
- "gunicorn"
|
|
- "-c"
|
|
- "/usr/local/etc/gunicorn/impress.py"
|
|
- "impress.wsgi:application"
|
|
- "--reload"
|
|
|
|
createsuperuser:
|
|
command:
|
|
- "/bin/sh"
|
|
- "-c"
|
|
- |
|
|
python manage.py createsuperuser --email admin@example.com --password admin
|
|
restartPolicy: Never
|
|
|
|
frontend:
|
|
envVars:
|
|
PORT: 8080
|
|
NEXT_PUBLIC_API_ORIGIN: https://impress.127.0.0.1.nip.io
|
|
|
|
replicas: 1
|
|
command:
|
|
- yarn
|
|
- dev
|
|
|
|
image:
|
|
repository: localhost:5001/impress-frontend
|
|
pullPolicy: Always
|
|
tag: "latest"
|
|
|
|
yProvider:
|
|
replicas: 1
|
|
|
|
image:
|
|
repository: localhost:5001/impress-y-provider
|
|
pullPolicy: Always
|
|
tag: "latest"
|
|
|
|
envVars:
|
|
COLLABORATION_LOGGING: true
|
|
COLLABORATION_SERVER_ORIGIN: https://impress.127.0.0.1.nip.io
|
|
COLLABORATION_SERVER_SECRET: my-secret
|
|
|
|
ingress:
|
|
enabled: true
|
|
host: impress.127.0.0.1.nip.io
|
|
|
|
ingressCollaborationWS:
|
|
enabled: true
|
|
host: impress.127.0.0.1.nip.io
|
|
|
|
annotations:
|
|
nginx.ingress.kubernetes.io/auth-url: https://impress.127.0.0.1.nip.io/api/v1.0/documents/collaboration-auth/
|
|
|
|
ingressCollaborationApi:
|
|
enabled: true
|
|
host: impress.127.0.0.1.nip.io
|
|
|
|
ingressAdmin:
|
|
enabled: true
|
|
host: impress.127.0.0.1.nip.io
|
|
|
|
ingressMedia:
|
|
enabled: true
|
|
host: impress.127.0.0.1.nip.io
|
|
|
|
annotations:
|
|
nginx.ingress.kubernetes.io/auth-url: https://impress.127.0.0.1.nip.io/api/v1.0/documents/media-auth/
|
|
nginx.ingress.kubernetes.io/auth-response-headers: "Authorization, X-Amz-Date, X-Amz-Content-SHA256"
|
|
nginx.ingress.kubernetes.io/upstream-vhost: minio.impress.svc.cluster.local:9000
|
|
nginx.ingress.kubernetes.io/rewrite-target: /impress-media-storage/$1
|
|
|
|
serviceMedia:
|
|
host: minio.impress.svc.cluster.local
|
|
port: 9000
|