inference

type API struct {
	// contains filtered or unexported fields
}

type CreateDeploymentRequest struct {
	// Region: region to target. If none is passed will use default region from the config.
	Region scw.Region `json:"-"`

	// Name: name of the deployment.
	Name string `json:"name"`

	// ProjectID: ID of the Project to create the deployment in.
	ProjectID string `json:"project_id"`

	// ModelID: ID of the model to use.
	ModelID string `json:"model_id"`

	// AcceptEula: if the model has an EULA, you must accept it before proceeding.
	// The terms of the EULA can be retrieved using the `GetModelEula` API call.
	AcceptEula *bool `json:"accept_eula,omitempty"`

	// NodeTypeName: name of the node type to use.
	NodeTypeName string `json:"node_type_name"`

	// Tags: list of tags to apply to the deployment.
	Tags []string `json:"tags"`

	// MinSize: defines the minimum size of the pool.
	MinSize *uint32 `json:"min_size,omitempty"`

	// MaxSize: defines the maximum size of the pool.
	MaxSize *uint32 `json:"max_size,omitempty"`

	// Endpoints: list of endpoints to create.
	Endpoints []*EndpointSpec `json:"endpoints"`

	// Quantization: quantization settings to apply to this deployment.
	Quantization *DeploymentQuantization `json:"quantization,omitempty"`
}

type CreateEndpointRequest struct {
	// Region: region to target. If none is passed will use default region from the config.
	Region scw.Region `json:"-"`

	// DeploymentID: ID of the deployment to create the endpoint for.
	DeploymentID string `json:"deployment_id"`

	// Endpoint: specification of the endpoint.
	Endpoint *EndpointSpec `json:"endpoint"`
}

type CreateModelRequest struct {
	// Region: region to target. If none is passed will use default region from the config.
	Region scw.Region `json:"-"`

	// Name: name of the model.
	Name string `json:"name"`

	// ProjectID: ID of the Project to import the model in.
	ProjectID string `json:"project_id"`

	// Source: where to import the model from.
	Source *ModelSource `json:"source"`
}

type DeleteDeploymentRequest struct {
	// Region: region to target. If none is passed will use default region from the config.
	Region scw.Region `json:"-"`

	// DeploymentID: ID of the deployment to delete.
	DeploymentID string `json:"-"`
}

type DeleteEndpointRequest struct {
	// Region: region to target. If none is passed will use default region from the config.
	Region scw.Region `json:"-"`

	// EndpointID: ID of the endpoint to delete.
	EndpointID string `json:"-"`
}

type DeleteModelRequest struct {
	// Region: region to target. If none is passed will use default region from the config.
	Region scw.Region `json:"-"`

	// ModelID: ID of the model to delete.
	ModelID string `json:"-"`
}

type Deployment struct {
	// ID: unique identifier.
	ID string `json:"id"`

	// Name: name of the deployment.
	Name string `json:"name"`

	// ProjectID: project ID.
	ProjectID string `json:"project_id"`

	// Status: status of the deployment.
	// Default value: unknown_status
	Status DeploymentStatus `json:"status"`

	// Tags: list of tags applied to the deployment.
	Tags []string `json:"tags"`

	// NodeTypeName: node type of the deployment.
	NodeTypeName string `json:"node_type_name"`

	// Endpoints: list of endpoints.
	Endpoints []*Endpoint `json:"endpoints"`

	// Size: current size of the pool.
	Size uint32 `json:"size"`

	// MinSize: defines the minimum size of the pool.
	MinSize uint32 `json:"min_size"`

	// MaxSize: defines the maximum size of the pool.
	MaxSize uint32 `json:"max_size"`

	// ErrorMessage: displays information if your deployment is in error state.
	ErrorMessage *string `json:"error_message"`

	// ModelID: ID of the model used for the deployment.
	ModelID string `json:"model_id"`

	// Quantization: quantization parameters for this deployment.
	Quantization *DeploymentQuantization `json:"quantization"`

	// ModelName: name of the deployed model.
	ModelName string `json:"model_name"`

	// CreatedAt: creation date of the deployment.
	CreatedAt *time.Time `json:"created_at"`

	// UpdatedAt: last modification date of the deployment.
	UpdatedAt *time.Time `json:"updated_at"`

	// Region: region of the deployment.
	Region scw.Region `json:"region"`
}

type DeploymentQuantization struct {
	// Bits: the number of bits each model parameter should be quantized to. The quantization method is chosen based on this value.
	Bits uint32 `json:"bits"`
}

type DeploymentStatus string

type Endpoint struct {
	// ID: unique identifier.
	ID string `json:"id"`

	// URL: for private endpoints, the URL will be accessible only from the Private Network.
	// In addition, private endpoints will expose a CA certificate that can be used to verify the server's identity.
	// This CA certificate can be retrieved using the `GetDeploymentCertificate` API call.
	URL string `json:"url"`

	// PublicNetwork: defines whether the endpoint is public.
	// Precisely one of PublicNetwork, PrivateNetwork must be set.
	PublicNetwork *EndpointPublicNetworkDetails `json:"public_network,omitempty"`

	// PrivateNetwork: details of the Private Network.
	// Precisely one of PublicNetwork, PrivateNetwork must be set.
	PrivateNetwork *EndpointPrivateNetworkDetails `json:"private_network,omitempty"`

	// DisableAuth: defines whether the authentication is disabled.
	DisableAuth bool `json:"disable_auth"`
}

type EndpointPrivateNetworkDetails struct {
	PrivateNetworkID string `json:"private_network_id"`
}

type EndpointPublicNetworkDetails struct{}

type EndpointSpec struct {
	// PublicNetwork: set the endpoint as public.
	// Precisely one of PublicNetwork, PrivateNetwork must be set.
	PublicNetwork *EndpointPublicNetworkDetails `json:"public_network,omitempty"`

	// PrivateNetwork: private endpoints are only accessible from the Private Network.
	// Precisely one of PublicNetwork, PrivateNetwork must be set.
	PrivateNetwork *EndpointPrivateNetworkDetails `json:"private_network,omitempty"`

	// DisableAuth: by default, deployments are protected by IAM authentication.
	// When setting this field to true, the authentication will be disabled.
	DisableAuth bool `json:"disable_auth"`
}

type GetDeploymentCertificateRequest struct {
	// Region: region to target. If none is passed will use default region from the config.
	Region scw.Region `json:"-"`

	DeploymentID string `json:"-"`
}

type GetDeploymentRequest struct {
	// Region: region to target. If none is passed will use default region from the config.
	Region scw.Region `json:"-"`

	// DeploymentID: ID of the deployment to get.
	DeploymentID string `json:"-"`
}

type GetModelRequest struct {
	// Region: region to target. If none is passed will use default region from the config.
	Region scw.Region `json:"-"`

	// ModelID: ID of the model to get.
	ModelID string `json:"-"`
}

type ListDeploymentsRequest struct {
	// Region: region to target. If none is passed will use default region from the config.
	Region scw.Region `json:"-"`

	// Page: page number to return.
	Page *int32 `json:"-"`

	// PageSize: maximum number of deployments to return per page.
	PageSize *uint32 `json:"-"`

	// OrderBy: order in which to return results.
	// Default value: created_at_desc
	OrderBy ListDeploymentsRequestOrderBy `json:"-"`

	// ProjectID: filter by Project ID.
	ProjectID *string `json:"-"`

	// OrganizationID: filter by Organization ID.
	OrganizationID *string `json:"-"`

	// Name: filter by deployment name.
	Name *string `json:"-"`

	// Tags: filter by tags.
	Tags []string `json:"-"`
}

type ListDeploymentsRequestOrderBy string

type ListDeploymentsResponse struct {
	// Deployments: list of deployments on the current page.
	Deployments []*Deployment `json:"deployments"`

	// TotalCount: total number of deployments.
	TotalCount uint64 `json:"total_count"`
}

type ListModelsRequest struct {
	// Region: region to target. If none is passed will use default region from the config.
	Region scw.Region `json:"-"`

	// OrderBy: order in which to return results.
	// Default value: display_rank_asc
	OrderBy ListModelsRequestOrderBy `json:"-"`

	// Page: page number to return.
	Page *int32 `json:"-"`

	// PageSize: maximum number of models to return per page.
	PageSize *uint32 `json:"-"`

	// ProjectID: filter by Project ID.
	ProjectID *string `json:"-"`

	// Name: filter by model name.
	Name *string `json:"-"`

	// Tags: filter by tags.
	Tags []string `json:"-"`
}

type ListModelsRequestOrderBy string

type ListModelsResponse struct {
	// Models: list of models on the current page.
	Models []*Model `json:"models"`

	// TotalCount: total number of models.
	TotalCount uint64 `json:"total_count"`
}

type ListNodeTypesRequest struct {
	// Region: region to target. If none is passed will use default region from the config.
	Region scw.Region `json:"-"`

	// Page: page number to return.
	Page *int32 `json:"-"`

	// PageSize: maximum number of node types to return per page.
	PageSize *uint32 `json:"-"`

	// IncludeDisabledTypes: include disabled node types in the response.
	IncludeDisabledTypes bool `json:"-"`
}

type ListNodeTypesResponse struct {
	// NodeTypes: list of node types.
	NodeTypes []*NodeType `json:"node_types"`

	// TotalCount: total number of node types.
	TotalCount uint64 `json:"total_count"`
}

type Model struct {
	// ID: unique identifier.
	ID string `json:"id"`

	// Name: unique Name identifier.
	Name string `json:"name"`

	// ProjectID: project ID.
	ProjectID string `json:"project_id"`

	// Tags: list of tags applied to the model.
	Tags []string `json:"tags"`

	// Status: status of the model.
	// Default value: unknown_status
	Status ModelStatus `json:"status"`

	// Description: purpose of the model.
	Description string `json:"description"`

	// ErrorMessage: displays information if your model is in error state.
	ErrorMessage *string `json:"error_message"`

	// HasEula: defines whether the model has an end user license agreement.
	HasEula bool `json:"has_eula"`

	// CreatedAt: creation date of the model.
	CreatedAt *time.Time `json:"created_at"`

	// UpdatedAt: last modification date of the model.
	UpdatedAt *time.Time `json:"updated_at"`

	// Region: region of the model.
	Region scw.Region `json:"region"`

	// NodesSupport: supported nodes types with quantization options and context lengths.
	NodesSupport []*ModelSupportInfo `json:"nodes_support"`

	// ParameterSizeBits: size, in bits, of the model parameters.
	ParameterSizeBits uint32 `json:"parameter_size_bits"`

	// SizeBytes: total size, in bytes, of the model files.
	SizeBytes uint64 `json:"size_bytes"`
}

type ModelSource struct {
	URL string `json:"url"`

	// Precisely one of Secret must be set.
	Secret *string `json:"secret,omitempty"`
}

type ModelStatus string

type ModelSupportInfo struct {
	// Nodes: list of supported node types.
	Nodes []*ModelSupportedNode `json:"nodes"`
}

type ModelSupportedNode struct {
	// NodeTypeName: supported node type.
	NodeTypeName string `json:"node_type_name"`

	// Quantizations: supported quantizations.
	Quantizations []*ModelSupportedQuantization `json:"quantizations"`
}

type ModelSupportedQuantization struct {
	// QuantizationBits: number of bits for this supported quantization.
	QuantizationBits uint32 `json:"quantization_bits"`

	// Allowed: tells whether this quantization is allowed for this node type.
	Allowed bool `json:"allowed"`

	// MaxContextSize: maximum inference context size available for this node type and quantization.
	MaxContextSize uint32 `json:"max_context_size"`
}

type NodeType struct {
	// Name: name of the node type.
	Name string `json:"name"`

	// StockStatus: current stock status for the node type.
	// Default value: unknown_stock
	StockStatus NodeTypeStock `json:"stock_status"`

	// Description: current specs of the offer.
	Description string `json:"description"`

	// Vcpus: number of virtual CPUs.
	Vcpus uint32 `json:"vcpus"`

	// Memory: quantity of RAM.
	Memory scw.Size `json:"memory"`

	// Vram: quantity of GPU RAM.
	Vram scw.Size `json:"vram"`

	// Disabled: the node type is currently disabled.
	Disabled bool `json:"disabled"`

	// Beta: the node type is currently in beta.
	Beta bool `json:"beta"`

	// CreatedAt: creation date of the node type.
	CreatedAt *time.Time `json:"created_at"`

	// UpdatedAt: last modification date of the node type.
	UpdatedAt *time.Time `json:"updated_at"`

	// Gpus: number of GPUs.
	Gpus uint32 `json:"gpus"`

	// Region: region of the node type.
	Region scw.Region `json:"region"`
}

type NodeTypeStock string

type UpdateDeploymentRequest struct {
	// Region: region to target. If none is passed will use default region from the config.
	Region scw.Region `json:"-"`

	// DeploymentID: ID of the deployment to update.
	DeploymentID string `json:"-"`

	// Name: name of the deployment.
	Name *string `json:"name,omitempty"`

	// Tags: list of tags to apply to the deployment.
	Tags *[]string `json:"tags,omitempty"`

	// MinSize: defines the new minimum size of the pool.
	MinSize *uint32 `json:"min_size,omitempty"`

	// MaxSize: defines the new maximum size of the pool.
	MaxSize *uint32 `json:"max_size,omitempty"`

	// ModelID: id of the model to set to the deployment.
	ModelID *string `json:"model_id,omitempty"`

	// Quantization: quantization to use to the deployment.
	Quantization *DeploymentQuantization `json:"quantization,omitempty"`
}

type UpdateEndpointRequest struct {
	// Region: region to target. If none is passed will use default region from the config.
	Region scw.Region `json:"-"`

	// EndpointID: ID of the endpoint to update.
	EndpointID string `json:"-"`

	// DisableAuth: by default, deployments are protected by IAM authentication.
	// When setting this field to true, the authentication will be disabled.
	DisableAuth *bool `json:"disable_auth,omitempty"`
}

type WaitForDeploymentRequest struct {
	DeploymentID  string
	Region        scw.Region
	Status        DeploymentStatus
	Timeout       *time.Duration
	RetryInterval *time.Duration
}

type WaitForModelRequest struct {
	ModelID       string
	Region        scw.Region
	Timeout       *time.Duration
	RetryInterval *time.Duration
}