Skip to main content

Module parallel

Module parallel 

Source
Expand description

Parallel optimization algorithms for multi-threaded training.

This module provides thread-safe optimizers that can leverage multiple CPU cores for parallel parameter updates, improving performance on multi-core systems.

§Key Features

  • Thread-Safe State Management: Lock-free and fine-grained locking strategies
  • Parallel Parameter Updates: Distribute parameter updates across threads
  • Work Stealing: Dynamic load balancing for uneven parameter distributions
  • NUMA Awareness: Optimize for Non-Uniform Memory Access architectures
  • Scalability: Efficient scaling from 2 to 64+ cores

Structs§

ParallelAdam
Parallel Adam optimizer with multi-threaded parameter updates.
ParallelConfig
Configuration for parallel optimization.
ParallelOptimizerState
Thread-safe optimizer state with fine-grained locking.
ParallelStats
Performance statistics for parallel optimization.
ParameterState
Individual parameter state with momentum and variance.

Traits§

BatchUpdate
Batch parameter update interface for better parallelization.