IRIS8 Talk to us

Platform · 03 Deploy

Three places your model can run

Your data centre, your cloud account, or cloud we manage in a region you choose. IRIS8 operates all three, and you own the weights.

Where it runs

Pick where it runs

The model is yours in every option. What changes is whose hardware and whose account.

Yours

Your buildingYour cloud accountThe region you pick

Yours

Your accountThe region you choose

Your apps
Your data

IRIS8 account, in the region you chooseSame region

Gateway
Your model
HardwareYour hardwareYour cloud provider, your accountBought and run by IRIS8

IRIS8 teamInside, scoped and logged accessInside the IRIS8 account

Your hardware, in your building. IRIS8 engineers work inside it with scoped, logged access.Your account with your cloud provider, in the region you pick. IRIS8 deploys into it and operates it.You own the weights. IRIS8 buys and runs the cloud capacity, in a region you choose.

Side by side

The three options, side by side

QuestionYour data centreYour cloud accountCloud we manage in your region
Who buys the hardware or holds the accountYou buy or lease the GPUsYou hold the account and pay the providerIRIS8 buys the capacity and holds the account
Where the model runsYour data centreYour cloud account, your regionAn IRIS8 account, in the region you choose
Where your data is processedIn your data centreIn your cloud accountIn the region you choose. Which data moves there is agreed in your contract
Network path from your appsInside your networkInside your cloud networkFrom your network to the IRIS8 account. Agreed in your contract
Who has accessYour staff. IRIS8 engineers, scoped and loggedYour staff. IRIS8 engineers, scoped and loggedIRIS8 engineers. Your access is agreed in your contract
Who operates itIRIS8IRIS8IRIS8
Who holds the weightsYou own them. They sit in your data centreYou own them. They sit in your accountYou own them. They sit in the IRIS8 account until handed to you
What you keep if you leaveThe model stays where it is, with its datasets, benchmarks and recordsThe model stays where it is, with its datasets, benchmarks and recordsWeights, datasets, benchmarks and records are handed to you. Agreed in your contract
Good fit whenYou have data centre capacity, or data must stay on your premisesYou already run on a major cloudYou have no GPU capacity or cloud team of your own

What happens

How a version is released

  • Hardware is sized in the first week
  • The model and the gateway are deployed together
  • A new version is released in stages, with the version pinned
  • The previous version is kept for rollback
IRIS8 console · loan-file-qa / releaseSample data

Release · loan-file-qa v3

Environment
Your cloud account
Stage 1
A small share of requests · Done
Stage 2
All requests · Live
Version
v3, pinned
Previous
v2, kept
Approved by
Head of Credit Operations

Roll back to v2

Your decisions

You decideIRIS8 does
Which of the three optionsSizes the hardware for the model and traffic
The regionDeploys the model and the gateway
Who on your side grants accessRuns the staged release
When a version goes liveKeeps rollback ready

Your decisions, our work

Take this with you

Pre-deployment checklist

For your infrastructure and security contacts, before week 2.

IRIS8 · Platform · Deploy1 page

Pre-deployment checklist

Where it runs

  • Option chosen: data centre, cloud account, or cloud managed by IRIS8
  • Region named, and the reason it must be that region
  • GPU capacity available or ordered, sized with IRIS8

Network

  • Which applications will call the model
  • Whether the model may reach the internet (default: no)
  • Whether any request may go to a frontier model, and for which tasks

Access

  • Named person who grants and removes IRIS8 engineer access
  • How access is requested, scoped and logged
  • Where the request log and model record will be stored

Data

  • Which fields are masked at the gateway
  • Which data may be used for training

Release

  • Named person who approves a version for go-live
  • Rollback tested before go-live

iris8.ai/platform/deploy.html

Questions

Questions

Talk to us
Q1Can we move from one option to another later?

Yes. You own the weights, so the model can be redeployed.

Q2Which clouds do you deploy on?

We confirm the cloud, region and GPU options for your setup in the first week. See the AWS and Azure pages for how we deploy in each.

Q3Can the model run with no internet access?

In your data centre or your cloud account, yes.

Q4Who pays for the infrastructure?

In your data centre or cloud account, you do, directly or passed through at cost. For cloud we manage, it is agreed in your contract.

Q5What hardware do we need on-prem?

GPUs sized to the model and the traffic. We size it in the first week, before you buy or reserve anything.

Next step

Tell us where it has to run