Module gradient 

Gradient and tensor storage with delta encoding

Provides efficient storage for neural network gradients and tensors:

• Delta encoding (store changes only)
• Sparse gradient compression
• Provenance metadata tracking
• Integration with version control

Example

use ipfrs_storage::{GradientStore, DeltaEncoder, SledBlockStore, BlockStoreConfig};
use std::sync::Arc;
use std::path::PathBuf;

// Create block store
let store = Arc::new(SledBlockStore::new(BlockStoreConfig {
    path: PathBuf::from(".ipfrs/gradients"),
    cache_size: 100 * 1024 * 1024,
})?);

// Create gradient store
let gradient_store = GradientStore::new(store);

// Store a gradient with delta encoding
let gradient = vec![1.0f32, 2.0, 3.0, 4.0];
let metadata = ProvenanceMetadata {
    layer: "layer1".to_string(),
    timestamp: 1234567890,
    training_config: "lr=0.001".to_string(),
};

let cid = gradient_store.store_gradient(&gradient, Some(metadata)).await?;

Structs

CompressionStats

Compression statistics

DeltaEncoder

Delta encoder for efficient gradient storage

GradientData

Gradient data with metadata

GradientStore

Gradient store for managing gradients with delta encoding

ProvenanceMetadata

Provenance metadata for tracking gradient origins