Struct MemoryEfficientTraining 

pub struct MemoryEfficientTraining;

Memory-efficient training utilities.

Implementations

impl MemoryEfficientTraining

pub fn optimal_batch_size( available_memory: usize, sample_size: usize, model_memory: usize, overhead_factor: f64, ) -> usize

Calculate optimal batch size for available memory.

Arguments

• available_memory - Available memory in bytes
• sample_size - Size of a single sample in bytes
• model_memory - Model memory footprint in bytes
• overhead_factor - Overhead multiplier (typically 2-4 for gradients)

pub fn estimate_model_memory( num_parameters: usize, with_gradients: bool, with_optimizer_state: bool, ) -> usize

Estimate memory for a model with given parameter count.

Arguments

• num_parameters - Number of model parameters
• with_gradients - Whether to include gradient storage
• with_optimizer_state - Whether to include optimizer state (e.g., Adam moments)

pub fn gradient_accumulation_steps( target_batch_size: usize, actual_batch_size: usize, ) -> usize

Calculate gradient accumulation steps for target batch size.

Arguments

• target_batch_size - Desired effective batch size
• actual_batch_size - Batch size that fits in memory

pub fn recommended_settings(gpu_memory_gb: f64) -> MemorySettings

Get recommended memory settings for a given GPU memory.