Struct DynamicBatchConfig 

pub struct DynamicBatchConfig {
    pub max_context_length: usize,
    pub avg_sentence_length: usize,
    pub min_contexts: usize,
    pub max_contexts: usize,
    pub same_document: bool,
    pub window_overlap: usize,
}

Dynamic batching configuration for cross-context training.

From xCoRe (Section 3.3): “At each step, we first sample the number of training contexts n in the range (1, ⌊w/s⌋), then construct a training batch by sampling n continuous contexts from d_i, with length equal to min(w, |d_i|)/n.”

This allows models to learn with both:

• Many small contexts (for cross-context learning)
• Few large contexts (for within-context quality)

Fields

max_context_length: usize

Maximum context length (tokens)

avg_sentence_length: usize

Average sentence length (for computing max contexts)

min_contexts: usize

Minimum number of contexts per batch

max_contexts: usize

Maximum number of contexts per batch

same_document: bool

Whether to sample contexts from same document (long-doc) or different docs (cross-doc)

window_overlap: usize

Overlap tokens between adjacent windows (for long-doc mode)

Implementations

impl DynamicBatchConfig

pub fn cross_document() -> Self

Create config for cross-document training.

pub fn long_document() -> Self

Create config for long-document training.

pub fn sample_num_contexts(&self, rng_seed: u64) -> usize

Compute the number of contexts to sample for this training step.

Uses uniform sampling in range (min_contexts, max_contexts).

pub fn context_length(&self, num_contexts: usize, doc_length: usize) -> usize

Compute context length given number of contexts.

From xCoRe: “length equal to min(w, |d_i|)/n”

Trait Implementations

impl Clone for DynamicBatchConfig

fn clone(&self) -> DynamicBatchConfig

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for DynamicBatchConfig

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

Formats the value using the given formatter.

impl Default for DynamicBatchConfig

fn default() -> Self

Returns the “default value” for a type.

impl<'de> Deserialize<'de> for DynamicBatchConfig

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

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Serialize for DynamicBatchConfig

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

where __S: Serializer,

Serialize this value into the given Serde serializer.