Kubernetes AI Toolchain Operator (Kaito)

What is NEW!
Latest Release: Nov 15th, 2024. Kaito v0.3.2.
First Release: Nov 15th, 2023. Kaito v0.1.0.

Kaito is an operator that automates the AI/ML model inference or tuning workload in a Kubernetes cluster. The target models are popular open-sourced large models such as falcon and phi-3. Kaito has the following key differentiations compared to most of the mainstream model deployment methodologies built on top of virtual machine infrastructures:

• Manage large model files using container images. A http server is provided to perform inference calls using the model library.
• Provide preset configurations to avoid adjusting workload parameters based on GPU hardware.
• Auto-provision GPU nodes based on model requirements.
• Host large model images in the public Microsoft Container Registry (MCR) if the license allows.

Using Kaito, the workflow of onboarding large AI inference models in Kubernetes is largely simplified.

Architecture

Kaito follows the classic Kubernetes Custom Resource Definition(CRD)/controller design pattern. User manages a workspace custom resource which describes the GPU requirements and the inference or tuning specification. Kaito controllers will automate the deployment by reconciling the workspace custom resource.

The above figure presents the Kaito architecture overview. Its major components consist of:

• Workspace controller: It reconciles the workspace custom resource, creates machine (explained below) custom resources to trigger node auto provisioning, and creates the inference or tuning workload (deployment, statefulset or job) based on the model preset configurations.
• Node provisioner controller: The controller's name is gpu-provisioner in gpu-provisioner helm chart. It uses the machine CRD originated from Karpenter to interact with the workspace controller. It integrates with Azure Resource Manager REST APIs to add new GPU nodes to the AKS cluster.

Note: The gpu-provisioner is an open sourced component. It can be replaced by other controllers if they support Karpenter-core APIs.

Installation

Please check the installation guidance here for deployment using Azure CLI and here for deployment using Terraform.

Quick start

After installing Kaito, one can try following commands to start a falcon-7b inference service.

$ cat examples/inference/kaito_workspace_falcon_7b.yaml
apiVersion: kaito.sh/v1alpha1
kind: Workspace
metadata:
  name: workspace-falcon-7b
resource:
  instanceType: "Standard_NC12s_v3"
  labelSelector:
    matchLabels:
      apps: falcon-7b
inference:
  preset:
    name: "falcon-7b"

$ kubectl apply -f examples/inference/kaito_workspace_falcon_7b.yaml

The workspace status can be tracked by running the following command. When the WORKSPACEREADY column becomes True, the model has been deployed successfully.

$ kubectl get workspace workspace-falcon-7b
NAME                  INSTANCE            RESOURCEREADY   INFERENCEREADY   WORKSPACEREADY   AGE
workspace-falcon-7b   Standard_NC12s_v3   True            True             True             10m

Next, one can find the inference service's cluster ip and use a temporal curl pod to test the service endpoint in the cluster.

$ kubectl get svc workspace-falcon-7b
NAME                  TYPE        CLUSTER-IP   EXTERNAL-IP   PORT(S)            AGE
workspace-falcon-7b   ClusterIP   <CLUSTERIP>  <none>        80/TCP,29500/TCP   10m

export CLUSTERIP=$(kubectl get svc workspace-falcon-7b -o jsonpath="{.spec.clusterIPs[0]}") 
$ kubectl run -it --rm --restart=Never curl --image=curlimages/curl -- curl -X POST http://$CLUSTERIP/chat -H "accept: application/json" -H "Content-Type: application/json" -d "{\"prompt\":\"YOUR QUESTION HERE\"}"

Usage

The detailed usage for Kaito supported models can be found in HERE. In case users want to deploy their own containerized models, they can provide the pod template in the inference field of the workspace custom resource (please see API definitions for details). The controller will create a deployment workload using all provisioned GPU nodes. Note that currently the controller does NOT handle automatic model upgrade. It only creates inference workloads based on the preset configurations if the workloads do not exist.

The number of the supported models in Kaito is growing! Please check this document to see how to add a new supported model.

Starting with version v0.3.0, Kaito supports model fine-tuning and using fine-tuned adapters in the inference service. Refer to the tuning document and inference document for more information.

FAQ

How do I ensure preferred nodes are correctly labeled for use in my workspace?

For using preferred nodes, make sure the node has the label specified in the labelSelector under matchLabels. For example, if your labelSelector is:

  labelSelector:
    matchLabels:
      apps: falcon-7b

Then the node should have the label: apps=falcon-7b.

How to upgrade the existing deployment to use the latest model configuration?

When using hosted public models, a user can delete the existing inference workload (Deployment of StatefulSet) manually, and the workspace controller will create a new one with the latest preset configuration (e.g., the image version) defined in the current release. For private models, it is recommended to create a new workspace with a new image version in the Spec.

How to update model/inference parameters to override the Kaito Preset Configuration?

Kaito provides a limited capability to override preset configurations for models that use transformer runtime manually. To update parameters for a deployed model, perform kubectl edit against the workload, which could be either a StatefulSet or Deployment. For example, to enable 4-bit quantization on a falcon-7b-instruct deployment, you would execute:

kubectl edit deployment workspace-falcon-7b-instruct

Within the deployment specification, locate and modify the command field.

Original

accelerate launch --num_processes 1 --num_machines 1 --machine_rank 0 --gpu_ids all inference_api.py --pipeline text-generation --torch_dtype bfloat16

Modify to enable 4-bit Quantization

accelerate launch --num_processes 1 --num_machines 1 --machine_rank 0 --gpu_ids all inference_api.py --pipeline text-generation --torch_dtype bfloat16 --load_in_4bit

Currently, we allow users to change the following paramenters manually:

• pipeline: For text-generation models this can be either text-generation or conversational.
• load_in_4bit or load_in_8bit: Model quantization resolution.

Should you need to customize other parameters, kindly file an issue for potential future inclusion.

What is the difference between instruct and non-instruct models?

The main distinction lies in their intended use cases. Instruct models are fine-tuned versions optimized for interactive chat applications. They are typically the preferred choice for most implementations due to their enhanced performance in conversational contexts. On the other hand, non-instruct, or raw models, are designed for further fine-tuning.

Contributing

Read more

This project welcomes contributions and suggestions. The contributions require you to agree to a Contributor License Agreement (CLA) declaring that you have the right to, and actually do, grant us the rights to use your contribution. For details, visit CLAs for CNCF.

When you submit a pull request, a CLA bot will automatically determine whether you need to provide a CLA and decorate the PR appropriately (e.g., status check, comment). Simply follow the instructions provided by the bot. You will only need to do this once across all repos using our CLA.

This project has adopted the CLAs for CNCF, please electronically sign the CLA via https://easycla.lfx.linuxfoundation.org. If you encounter issues, you can submit a ticket with the Linux Foundation ID group through the Linux Foundation Support website.

License

See MIT License.

Code of Conduct

KAITO has adopted the Cloud Native Compute Foundation Code of Conduct. For more information see the KAITO Code of Conduct.

Contact

"Kaito devs" [email protected]