Struct StepStats 

pub struct StepStats {
    pub loss: f32,
    pub tokens: usize,
    pub optimizer_step: bool,
    pub grad_norm: Option<f32>,
    pub lr: Option<f32>,
}

Result of one Trainer::train_step_masked call.

Fields

loss: f32

Mean cross-entropy over the supervised tokens of this sequence.

tokens: usize

Number of supervised (unmasked) target tokens.

optimizer_step: bool

Whether this call completed an accumulation window and updated the parameters (false while accumulating or after an fp16 overflow).

grad_norm: Option<f32>

Global L2 norm of the gradients before clipping, when the optimizer stepped.

lr: Option<f32>

Learning rate used for the optimizer step.

Trait Implementations

impl Clone for StepStats

fn clone(&self) -> Self

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Copy for StepStats

impl Debug for StepStats

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

Formats the value using the given formatter.

impl PartialEq for StepStats

fn eq(&self, other: &Self) -> bool

Equality operator ==.

fn ne(&self, other: &Rhs) -> bool

Inequality operator !=.

impl StructuralPartialEq for StepStats