Module tensor 

Core gradient tensor types and operations.

This module provides the fundamental gradient data structures:

• SparseGradient — sparse representation with index/value pairs
• QuantizedGradient — int8-quantized representation for compression
• LayerGradient — enum over dense / sparse / quantized
• GradientDelta — per-layer gradient bundle linked to a base model CID
• GradientCompressor — top-k, threshold, random, and quantization compression
• GradientAggregator — average, weighted-average, and momentum helpers
• GradientVerifier — shape, finiteness, outlier, clipping utilities

Structs

GradientAggregator

Gradient aggregation for federated learning

GradientCompressor

Gradient compression utilities

GradientDelta

Gradient delta (difference from base model)

GradientVerifier

Gradient verification utilities

QuantizedGradient

Quantized gradient (reduced precision)

SparseGradient

Sparse gradient representation

Enums

LayerGradient

Gradient for a single layer