Struct TrainingConfig 

pub struct TrainingConfig {

    pub device_config: DeviceConfig,
    pub learning_rate: f64,
    pub batch_size: usize,
    pub epochs: usize,
    pub weight_decay: f64,
    pub grad_clip: Option<f32>,
    pub beta1: f64,
    pub beta2: f64,
    pub eps: f64,
    pub scheduler: Option<SchedulerType>,
    pub track_metrics: bool,
    pub log_interval: usize,
    pub validation_split: f32,
    pub early_stopping_patience: Option<usize>,
    pub use_gradient_checkpointing: bool,
    pub checkpoint_segment_size: Option<usize>,
    pub mixed_precision: MixedPrecision,
    pub loss_scale: f32,
}

Configuration for training

Fields

device_config: DeviceConfig

Device configuration (CPU/CUDA/Metal)

learning_rate: f64

Learning rate (initial for schedulers)

batch_size: usize

Batch size

epochs: usize

Number of epochs

weight_decay: f64

Weight decay (L2 regularization)

grad_clip: Option<f32>

Gradient clipping threshold

beta1: f64

Beta1 for Adam optimizer

beta2: f64

Beta2 for Adam optimizer

eps: f64

Epsilon for Adam optimizer

scheduler: Option<SchedulerType>

Learning rate scheduler type

track_metrics: bool

Enable metrics tracking

log_interval: usize

Log interval (batches)

validation_split: f32

Validation split (0.0 to 1.0)

early_stopping_patience: Option<usize>

Early stopping patience (epochs)

use_gradient_checkpointing: bool

Enable gradient checkpointing (saves memory by recomputing activations)

checkpoint_segment_size: Option<usize>

Checkpoint every N layers (None = checkpoint all layers)

mixed_precision: MixedPrecision

Mixed precision training mode

loss_scale: f32

Loss scaling factor for mixed precision (to prevent underflow)

Implementations

impl TrainingConfig

pub fn with_scheduler(self, scheduler: SchedulerType) -> Self

Set scheduler type

pub fn without_metrics(self) -> Self

Disable metrics tracking

pub fn with_validation_split(self, split: f32) -> Self

Set validation split

pub fn with_early_stopping(self, patience: usize) -> Self

Set early stopping patience

pub fn without_early_stopping(self) -> Self

Disable early stopping

pub fn with_gradient_checkpointing(self, segment_size: Option<usize>) -> Self

Enable gradient checkpointing for memory-efficient training

pub fn without_gradient_checkpointing(self) -> Self

Disable gradient checkpointing

pub fn with_fp16(self) -> Self

Enable mixed precision training (FP16)

pub fn with_bf16(self) -> Self

Enable mixed precision training (BF16)

pub fn with_mixed_precision(self, mode: MixedPrecision, loss_scale: f32) -> Self

Set mixed precision mode

pub fn without_mixed_precision(self) -> Self

Disable mixed precision training

Trait Implementations

impl Clone for TrainingConfig

fn clone(&self) -> TrainingConfig

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for TrainingConfig

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Default for TrainingConfig

fn default() -> Self

Returns the “default value” for a type.

impl<'de> Deserialize<'de> for TrainingConfig

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Serialize for TrainingConfig

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.