Module gradient 

Gradient storage and management for federated learning

This module provides:

• Gradient delta format (differences from base model)
• Gradient compression (sparsification, quantization, top-k)
• Gradient aggregation (averaging, weighted, momentum)
• Gradient verification (checksum, shape, outliers)
• Backward pass coordination via TensorSwap
• CID-linked computation graph for gradient tracking
• Gradient checkpointing for training resumption

Re-exports

pub use arrow_ipc::load_gradient_from_arrow;

pub use arrow_ipc::store_gradient_as_arrow;

pub use backward_pass::clip_gradient_norm;

pub use backward_pass::federated_average;

pub use backward_pass::AggregationMethod;

pub use backward_pass::BackwardPassConfig;

pub use backward_pass::BackwardPassCoordinator as LegacyBackwardPassCoordinator;

pub use backward_pass::BackwardPassStats;

pub use backward_pass::BackwardPassStep;

pub use backward_pass::BackwardStepStatus;

pub use checkpoint::GradientCheckpoint;

pub use computation_graph::ComputationGraphError;

pub use computation_graph::ComputationGraphStore;

pub use computation_graph::ComputationNode;

pub use federated::ClientInfo;

pub use federated::ClientState;

pub use federated::ConvergenceConfig;

pub use federated::ConvergenceDetector;

pub use federated::DPMechanism;

pub use federated::DifferentialPrivacy;

pub use federated::DistributedGradientAccumulator;

pub use federated::FederatedError;

pub use federated::FederatedRound;

pub use federated::GossipModelSync;

pub use federated::ModelSyncProtocol;

pub use federated::ModelUpdate;

pub use federated::PrivacyBudget;

pub use federated::RoundStats;

pub use federated::SecureAggregation;

pub use tensor::GradientAggregator;

pub use tensor::GradientCompressor;

pub use tensor::GradientDelta;

pub use tensor::GradientVerifier;

pub use tensor::LayerGradient;

pub use tensor::QuantizedGradient;

pub use tensor::SparseGradient;

pub use coordination::ArrowBlockError;

pub use coordination::BackwardPassCoordinator;

pub use coordination::BackwardPassId;

pub use coordination::CoordinationError;

pub use coordination::CoordinationStatus;

pub use coordination::GradientArrowBlock;

pub use coordination::GradientContribution;

Modules

arrow_ipc

Standalone Arrow IPC helpers for gradient serialization.

backward_pass

Backward pass coordination via TensorSwap.

checkpoint

Gradient checkpointing — snapshot and restore training state.

computation_graph

CID-linked computation graph for gradient tracking.

coordination

Gradient coordination protocol for distributed backward passes.

federated

Federated learning utilities.

tensor

Core gradient tensor types and operations.

Enums

GradientError

Errors that can occur during gradient operations