SparseUpdate<T>
Models & Types · AiDotNet.FederatedLearning.Adapters
Represents a sparse update: only a subset of indices have non-zero values. This is the communication-efficient representation used by SparseLoRA.
Constructors
| Constructor | Summary |
|---|---|
SparseUpdate(int[] indices, T[] values, int totalLength) |
Creates a new sparse update. |
Properties
| Property | Summary |
|---|---|
T[] Values { get; } |
Values at the non-zero indices. |
double Density { get; } |
Effective sparsity: fraction of elements that are non-zero. |
int NnzCount { get; } |
Number of non-zero elements. |
int TotalLength { get; } |
Total length of the full dense vector. |
int[] Indices { get; } |
Indices of non-zero elements. |