Enum TrainingRegime 

pub enum TrainingRegime {
    Finetune,
    FromScratch {
        vocab_size: u32,
    },
}

Training regime selects which epoch-zero val_loss cap INV-TRAIN-005 enforces. The doubling rule at N ≥ 1 is identical across regimes.

Contract: training-loop-pretrain-v1.yaml v1.2.0 non_divergence.

• TrainingRegime::Finetune — base weights pretrained; epoch-zero cap is 10.0, matching the MODEL-1 literal.
• TrainingRegime::FromScratch — random init; epoch-zero cap is 2.0 × ln(vocab_size), i.e. 2× the uniform-random cross-entropy baseline. For vocab=50257 the cap is ≈21.64.

Variants

Finetune

FromScratch

Fields

vocab_size: u32

Implementations

impl TrainingRegime

pub fn epoch_zero_val_loss_limit(&self) -> f32

Regime-dependent cap on val_loss[0] — the v1.2.0 amendment.

Finetune: EPOCH_ZERO_VAL_LOSS_LIMIT (10.0, MODEL-1 literal). FromScratch: DIVERGENCE_RATIO_LIMIT × ln(vocab_size). For vocab_size=50257 this is ≈21.64. vocab_size < 2 is clamped to 2 so ln stays positive — operator misuse, not worth a panic.

Trait Implementations

impl Clone for TrainingRegime

fn clone(&self) -> TrainingRegime

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Copy for TrainingRegime

impl Debug for TrainingRegime

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

Formats the value using the given formatter.

impl Default for TrainingRegime

fn default() -> Self

Returns the “default value” for a type.

impl<'de> Deserialize<'de> for TrainingRegime

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

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl PartialEq for TrainingRegime

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

Tests for self and other values to be equal, and is used by ==.

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

Tests for !=. The default implementation is almost always sufficient, and should not be overridden without very good reason.

impl Serialize for TrainingRegime

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

where __S: Serializer,

Serialize this value into the given Serde serializer.

impl StructuralPartialEq for TrainingRegime