Skip to content

API Reference ​

Packages ​

thalamus.cloud/v1alpha1 ​

Package v1alpha1 contains API Schema definitions for the thalamus.cloud v1alpha1 API group.

Resource Types ​

BackendType ​

Underlying type: string

Validation:

  • Enum: [native kserve kaito]

Appears in:

FieldDescription
native
kserve
kaito

EPPType ​

Underlying type: string

Validation:

  • Enum: [llm-d unknown]

Appears in:

FieldDescription
llm-d
unknown

EngineType ​

Underlying type: string

Validation:

  • Enum: [vllm unknown]

Appears in:

FieldDescription
vllm
unknown

HFWeightsSpec ​

HFWeightsSpec configures model weights sourced from Hugging Face.

Appears in:

FieldDescriptionDefaultValidation
repoId stringRepoID is the Hugging Face repository ID, e.g. "openai/gpt-oss-120b".
tokenSecret SecretKeySelectorTokenSecret references the key within a Kubernetes secret containing the Hugging Face token.Optional: {}

Model ​

Model is the Schema for the models API.

Appears in:

FieldDescriptionDefaultValidation
apiVersion stringthalamus.cloud/v1alpha1
kind stringModel
metadata ObjectMetaRefer to Kubernetes API documentation for fields of metadata.Optional: {}
spec ModelSpecspec defines the desired state of the Model.Required: {}
status ModelStatusstatus defines the observed state of the Model.Optional: {}

ModelList ​

ModelList contains a list of Model.

FieldDescriptionDefaultValidation
apiVersion stringthalamus.cloud/v1alpha1
kind stringModelList
metadata ListMetaRefer to Kubernetes API documentation for fields of metadata.
items Model array

ModelPhase ​

Underlying type: string

Appears in:

FieldDescription
PendingTODO: set Pending when admission/resource checks block reconciliation (e.g. insufficient GPU quota). Currently unused.
Creating
Ready
Inactive
Failed

ModelSpec ​

ModelSpec defines the configuration for the model instance.

Appears in:

FieldDescriptionDefaultValidation
backend BackendTypeBackend selects which inference backend manages this model's resources.
Defaults to "native".
nativeEnum: [native kserve kaito]
Optional: {}
replicas integerReplicas defines the number of engine replicas to deploy for this model.
Defaults to 1.
1Minimum: 0
Optional: {}
serving ServingSpecServing defines how the model is served.
weights WeightsSpecWeights defines where the model weights are sourced from.
aliases string arrayAliases are the model names under which this model is exposed.
When empty, the base model name from the weights is used.
MaxItems: 5
Optional: {}
items:MaxLength: 4096
items:MinLength: 1
items:Pattern: ^[!-~]+([\t ]?[!-~]+)*$

ModelStatus ​

ModelStatus defines the observed state of a Model.

Appears in:

FieldDescriptionDefaultValidation
engineType EngineTypeEngineType is the detected inference engine type, derived from the engine image.Enum: [vllm unknown]
Optional: {}
eppType EPPTypeEPPType is the detected EPP type, derived from the EPP image.Enum: [llm-d unknown]
Optional: {}
phase ModelPhasePhase summarizes the current lifecycle state of the model.Optional: {}
conditions Condition arrayConditions represent the latest observations of the model's state.Optional: {}

ServingSpec ​

ServingSpec defines the serving configuration for the model instance.

Appears in:

FieldDescriptionDefaultValidation
engine PodSpecEngine is the pod spec for the inference engine. The container named
'engine' is the main container and must set an image. Containers with
other names are added as sidecars.
router PodSpecRouter is the pod spec for the entry point. The container named
'router' is the main container and must set an image. Containers with
other names are added as sidecars.

WeightsSpec ​

WeightsSpec defines where the model weights are sourced from.

Appears in:

FieldDescriptionDefaultValidation
type WeightsTypeType of the weights source.Enum: [hf]
hf HFWeightsSpecHF configures weights sourced from Hugging Face.Optional: {}

WeightsType ​

Underlying type: string

Validation:

  • Enum: [hf]

Appears in:

FieldDescription
hf