Enum PolicyBroadcast 

pub enum PolicyBroadcast {
    Bytes {
        version: u64,
        bytes: Arc<Vec<u8>>,
    },
}

In-process policy broadcast from a learner to its actors.

Cross-thread, in-process counterpart to PolicyUpdate (which publishes a saved-model file path). This enum is what the single-host asynchronous actor-learner runner (crate::train::ppo::actor_learner) sends over each per-actor crossbeam_channel broadcast channel after a learner update. The payload representation is chosen by the learner implementation; the actor side stays agnostic by matching on the variant.

Variants

Bytes

Serialized Burn module record bytes, produced by burn::record::BinBytesRecorder with burn::record::FullPrecisionSettings.

Bytes are always Send regardless of the tensor backend (unlike raw module clones, whose Send-ness depends on the backend’s tensor primitives), and the Arc keeps the N-actor fan-out allocation-free — each actor channel receives a cheap pointer clone of the same serialized blob.

Fields

version: u64

Monotonically increasing policy version. Incremented by the learner once per broadcast so actors (and tests) can observe how fresh their local policy copy is.

bytes: Arc<Vec<u8>>

Shared serialized module record.

Implementations

impl PolicyBroadcast

pub fn version(&self) -> u64

The policy version carried by this broadcast.

Trait Implementations

impl Clone for PolicyBroadcast

fn clone(&self) -> PolicyBroadcast

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for PolicyBroadcast

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

Formats the value using the given formatter.