skainet-lang-ksp-annotations/sk.ainet.lang.ops

Package-level declarations

Types

@Target(allowedTargets = [AnnotationTarget.CLASS])

annotation class Backend(val id: String, val displayName: String = "", val internal: Boolean = false)

Marks a class as a concrete compute backend implementation of the TensorOps interface. The docs KSP processor uses this to derive the statusByBackend map for each operator automatically, so adding a new backend is one annotation instead of N hand-edits to @InProgress.

ComputationMode

enum ComputationMode : Enum<ComputationMode>

Computation mode for the Mikrograd annotation. This determines whether to use ForwardValue (INFERENCE) or BackwardValue (TRAINING).

DslOp

@Target(allowedTargets = [AnnotationTarget.FUNCTION])

annotation class DslOp(val category: String = "", val description: String = "")

Annotation for functions that are part of the SKaiNET DSL. These are typically composite operations built from primitive tensor operations.

InProgress

@Target(allowedTargets = [AnnotationTarget.CLASS, AnnotationTarget.FUNCTION])

annotation class InProgress(val backends: String, val owner: String = "", val issue: String = "")

Annotation to mark classes or functions as in progress for specific backends.

NotImplemented

@Target(allowedTargets = [AnnotationTarget.CLASS, AnnotationTarget.FUNCTION])

annotation class NotImplemented(val backends: String)

Annotation to mark classes or functions as not implemented for specific backends.

TensorOp

@Target(allowedTargets = [AnnotationTarget.FUNCTION])

annotation class TensorOp(val mode: ComputationMode = ComputationMode.INFERENCE)

Annotation for functions that should be processed by the Mikrograd KSP processor. The processor will generate optimized code for the function based on the computation mode.