skainet-io-core/sk.ainet.io.weights/WeightTensor

WeightTensor

data class WeightTensor<T : DType, V>(val name: String, val shape: List<Int>, val tensor: Tensor<T, V>, val isBias: Boolean = name.lowercase().contains("bias"))(source)

Format-agnostic representation of a weight tensor loaded from any model format (ONNX, GGUF, SafeTensors, etc.).

This abstraction allows the weight mapping logic to be decoupled from the specific serialization format, enabling code reuse across different loaders.

Constructors

WeightTensor

constructor(name: String, shape: List<Int>, tensor: Tensor<T, V>, isBias: Boolean = name.lowercase().contains("bias"))

Properties

isBias

val isBias: Boolean

Whether this tensor represents a bias parameter (auto-detected from name)

name

val name: String

The original name of the tensor as stored in the model file

shape

val shape: List<Int>

The dimensions of the tensor

tensor

val tensor: Tensor<T, V>

The actual tensor data