Documentation
¶
Overview ¶
Package v1 contains API Schema definitions for the kubeai v1 API group +kubebuilder:object:generate=true +groupName=kubeai.org
Index ¶
Constants ¶
const ( PodModelLabel = "model" // PodHashLabel is a label key used to store the hash of the Pod spec // that was used to create the Pod. This is used to determine if a Pod // needs to be recreated. PodHashLabel = "pod-hash" ModelFeatureLabelDomain = "features.kubeai.org" // ModelPodIPAnnotation is the annotation key used to specify an IP // to use for the model Pod instead of the IP address in the status of the Pod. // Use in conjunction with --allow-pod-address-override for development purposes. ModelPodIPAnnotation = "model-pod-ip" ModelPodPortAnnotation = "model-pod-port" ModelCacheEvictionFinalizer = "kubeai.org/cache-eviction" )
const ( ModelFeatureTextGeneration = "TextGeneration" ModelFeatureTextEmbedding = "TextEmbedding" // TODO (samos123): Add validation that Speech to Text only supports Faster Whisper. ModelFeatureSpeechToText = "SpeechToText" )
const ( OLlamaEngine = "OLlama" VLLMEngine = "VLLM" FasterWhisperEngine = "FasterWhisper" InfinityEngine = "Infinity" )
const (
PodAdapterLabelPrefix = "adapter.kubeai.org/"
)
Variables ¶
var ( // GroupVersion is group version used to register these objects GroupVersion = schema.GroupVersion{Group: "kubeai.org", Version: "v1"} // SchemeBuilder is used to add go types to the GroupVersionKind scheme SchemeBuilder = &scheme.Builder{GroupVersion: GroupVersion} // AddToScheme adds the types in this group-version to the given scheme. AddToScheme = SchemeBuilder.AddToScheme )
Functions ¶
func PVCModelAnnotation ¶ added in v0.9.0
func PodAdapterLabel ¶ added in v0.11.0
Types ¶
type Adapter ¶ added in v0.11.0
type Adapter struct {
// Name must be a lowercase string with no spaces.
// +kubebuilder:validation:Required
// +kubebuilder:validation:Pattern=^[a-z0-9-]+$
// +kubebuilder:validation:MaxLength=63
Name string `json:"name"`
// +kubebuilder:validation:XValidation:rule="self.startsWith(\"hf://\") || self.startsWith(\"s3://\") || self.startsWith(\"gs://\") || self.startsWith(\"oss://\")", message="adapter url must start with \"hf://\", \"s3://\", \"gs://\", or \"oss://\"."
URL string `json:"url"`
}
func (*Adapter) DeepCopy ¶ added in v0.11.0
DeepCopy is an autogenerated deepcopy function, copying the receiver, creating a new Adapter.
func (*Adapter) DeepCopyInto ¶ added in v0.11.0
DeepCopyInto is an autogenerated deepcopy function, copying the receiver, writing into out. in must be non-nil.
type Model ¶
type Model struct {
metav1.TypeMeta `json:",inline"`
metav1.ObjectMeta `json:"metadata,omitempty"`
Spec ModelSpec `json:"spec,omitempty"`
Status ModelStatus `json:"status,omitempty"`
}
Model resources define the ML models that will be served by KubeAI. +kubebuilder:object:root=true +kubebuilder:subresource:status +kubebuilder:subresource:scale:specpath=.spec.replicas,statuspath=.status.replicas.all +kubebuilder:validation:XValidation:rule="size(self.metadata.name) <= 40", message="name must not exceed 40 characters."
func (*Model) DeepCopy ¶
DeepCopy is an autogenerated deepcopy function, copying the receiver, creating a new Model.
func (*Model) DeepCopyInto ¶
DeepCopyInto is an autogenerated deepcopy function, copying the receiver, writing into out. in must be non-nil.
func (*Model) DeepCopyObject ¶
DeepCopyObject is an autogenerated deepcopy function, copying the receiver, creating a new runtime.Object.
type ModelFeature ¶
type ModelFeature string
+kubebuilder:validation:Enum=TextGeneration;TextEmbedding;SpeechToText
type ModelList ¶
type ModelList struct {
metav1.TypeMeta `json:",inline"`
metav1.ListMeta `json:"metadata,omitempty"`
Items []Model `json:"items"`
}
ModelList contains a list of Models.
func (*ModelList) DeepCopy ¶
DeepCopy is an autogenerated deepcopy function, copying the receiver, creating a new ModelList.
func (*ModelList) DeepCopyInto ¶
DeepCopyInto is an autogenerated deepcopy function, copying the receiver, writing into out. in must be non-nil.
func (*ModelList) DeepCopyObject ¶
DeepCopyObject is an autogenerated deepcopy function, copying the receiver, creating a new runtime.Object.
type ModelSpec ¶
type ModelSpec struct {
// URL of the model to be served.
// Currently the following formats are supported:
//
// For VLLM, FasterWhisper, Infinity engines:
//
// "hf://<repo>/<model>"
// "pvc://<pvcName>"
// "pvc://<pvcName>/<pvcSubpath>"
// "gs://<bucket>/<path>" (only with cacheProfile)
// "oss://<bucket>/<path>" (only with cacheProfile)
// "s3://<bucket>/<path>" (only with cacheProfile)
//
// For OLlama engine:
//
// "ollama://<model>"
//
// +kubebuilder:validation:Required
// +kubebuilder:validation:XValidation:rule="self == oldSelf", message="url is immutable."
// +kubebuilder:validation:XValidation:rule="self.startsWith(\"hf://\") || self.startsWith(\"pvc://\") || self.startsWith(\"ollama://\") || self.startsWith(\"s3://\") || self.startsWith(\"gs://\") || self.startsWith(\"oss://\")", message="url must start with \"hf://\", \"pvc://\", \"ollama://\", \"s3://\", \"gs://\", or \"oss://\" and not be empty."
URL string `json:"url"`
Adapters []Adapter `json:"adapters,omitempty"`
// Features that the model supports.
// Dictates the APIs that are available for the model.
Features []ModelFeature `json:"features"`
// Engine to be used for the server process.
// +kubebuilder:validation:Enum=OLlama;VLLM;FasterWhisper;Infinity
// +kubebuilder:validation:Required
Engine string `json:"engine"`
// ResourceProfile required to serve the model.
// Use the format "<resource-profile-name>:<count>".
// Example: "nvidia-gpu-l4:2" - 2x NVIDIA L4 GPUs.
// Must be a valid ResourceProfile defined in the system config.
ResourceProfile string `json:"resourceProfile,omitempty"`
// CacheProfile to be used for caching model artifacts.
// Must be a valid CacheProfile defined in the system config.
// +kubebuilder:validation:XValidation:rule="self == oldSelf", message="cacheProfile is immutable."
CacheProfile string `json:"cacheProfile,omitempty"`
// Image to be used for the server process.
// Will be set from ResourceProfile + Engine if not specified.
Image string `json:"image,omitempty"`
// Args to be added to the server process.
Args []string `json:"args,omitempty"`
// Env variables to be added to the server process.
Env map[string]string `json:"env,omitempty"`
// Replicas is the number of Pod replicas that should be actively
// serving the model. KubeAI will manage this field unless AutoscalingDisabled
// is set to true.
Replicas *int32 `json:"replicas,omitempty"`
// MinReplicas is the minimum number of Pod replicas that the model can scale down to.
// Note: 0 is a valid value.
// +kubebuilder:validation:Minimum=0
// +kubebuilder:validation:Optional
MinReplicas int32 `json:"minReplicas"`
// MaxReplicas is the maximum number of Pod replicas that the model can scale up to.
// Empty value means no limit.
// +kubebuilder:validation:Minimum=1
MaxReplicas *int32 `json:"maxReplicas,omitempty"`
// AutoscalingDisabled will stop the controller from managing the replicas
// for the Model. When disabled, metrics will not be collected on server Pods.
AutoscalingDisabled bool `json:"autoscalingDisabled,omitempty"`
// TargetRequests is average number of active requests that the autoscaler
// will try to maintain on model server Pods.
// +kubebuilder:validation:Minimum=1
// +kubebuilder:default=100
TargetRequests *int32 `json:"targetRequests"`
// ScaleDownDelay is the minimum time before a deployment is scaled down after
// the autoscaling algorithm determines that it should be scaled down.
// +kubebuilder:default=30
ScaleDownDelaySeconds *int64 `json:"scaleDownDelaySeconds"`
// Owner of the model. Used solely to populate the owner field in the
// OpenAI /v1/models endpoint.
// DEPRECATED.
// +kubebuilder:validation:Optional
Owner string `json:"owner"`
}
ModelSpec defines the desired state of Model. +kubebuilder:validation:XValidation:rule="!has(self.cacheProfile) || self.url.startsWith(\"hf://\") || self.url.startsWith(\"s3://\") || self.url.startsWith(\"gs://\") || self.url.startsWith(\"oss://\")", message="cacheProfile is only supported with urls of format \"hf://...\", \"s3://...\", \"gs://...\", or \"oss://...\" at the moment." +kubebuilder:validation:XValidation:rule="!self.url.startsWith(\"s3://\") || has(self.cacheProfile)", message="urls of format \"s3://...\" only supported when using a cacheProfile" +kubebuilder:validation:XValidation:rule="!self.url.startsWith(\"gs://\") || has(self.cacheProfile)", message="urls of format \"gs://...\" only supported when using a cacheProfile" +kubebuilder:validation:XValidation:rule="!self.url.startsWith(\"oss://\") || has(self.cacheProfile)", message="urls of format \"oss://...\" only supported when using a cacheProfile" +kubebuilder:validation:XValidation:rule="!has(self.maxReplicas) || self.minReplicas <= self.maxReplicas", message="minReplicas should be less than or equal to maxReplicas." +kubebuilder:validation:XValidation:rule="!has(self.adapters) || self.engine == \"VLLM\"", message="adapters only supported with VLLM engine."
func (*ModelSpec) DeepCopy ¶
DeepCopy is an autogenerated deepcopy function, copying the receiver, creating a new ModelSpec.
func (*ModelSpec) DeepCopyInto ¶
DeepCopyInto is an autogenerated deepcopy function, copying the receiver, writing into out. in must be non-nil.
type ModelStatus ¶
type ModelStatus struct {
Replicas ModelStatusReplicas `json:"replicas,omitempty"`
Cache *ModelStatusCache `json:"cache,omitempty"`
}
ModelStatus defines the observed state of Model.
func (*ModelStatus) DeepCopy ¶
func (in *ModelStatus) DeepCopy() *ModelStatus
DeepCopy is an autogenerated deepcopy function, copying the receiver, creating a new ModelStatus.
func (*ModelStatus) DeepCopyInto ¶
func (in *ModelStatus) DeepCopyInto(out *ModelStatus)
DeepCopyInto is an autogenerated deepcopy function, copying the receiver, writing into out. in must be non-nil.
type ModelStatusCache ¶ added in v0.9.0
type ModelStatusCache struct {
Loaded bool `json:"loaded"`
}
func (*ModelStatusCache) DeepCopy ¶ added in v0.9.0
func (in *ModelStatusCache) DeepCopy() *ModelStatusCache
DeepCopy is an autogenerated deepcopy function, copying the receiver, creating a new ModelStatusCache.
func (*ModelStatusCache) DeepCopyInto ¶ added in v0.9.0
func (in *ModelStatusCache) DeepCopyInto(out *ModelStatusCache)
DeepCopyInto is an autogenerated deepcopy function, copying the receiver, writing into out. in must be non-nil.
type ModelStatusReplicas ¶
func (*ModelStatusReplicas) DeepCopy ¶
func (in *ModelStatusReplicas) DeepCopy() *ModelStatusReplicas
DeepCopy is an autogenerated deepcopy function, copying the receiver, creating a new ModelStatusReplicas.
func (*ModelStatusReplicas) DeepCopyInto ¶
func (in *ModelStatusReplicas) DeepCopyInto(out *ModelStatusReplicas)
DeepCopyInto is an autogenerated deepcopy function, copying the receiver, writing into out. in must be non-nil.
Source Files