burn_p2p

burn_p2p turns a burn learner into a peer in a decentralized training network.

core shape:

trainers download the current head and assigned data shard, run one train window, then publish a candidate head
validators evaluate candidates and promote accepted updates
the same network can include native peers, browser peers, viewers, and validators

install

[dependencies]
burn_p2p = { version = "=0.21.0-pre.5", features = ["burn"] }

happy path

use burn::train::Learner;
use burn_p2p::{ExperimentId, RevisionId, StorageConfig, StudyId};
use burn_p2p::burn::from_loaders;

let mut trainer = from_loaders(
    Learner::new(model, optimizer, scheduler),
    device,
    train_loader,
    eval_loader,
)
.trainer(release_manifest, supported_workload)?
.with_network(network_manifest)?
.with_storage(StorageConfig::new("./node"))
.with_bootstrap_peer(validator_addr)
.spawn()?;

let experiment = trainer.experiment(
    StudyId::new("study"),
    ExperimentId::new("mnist"),
    RevisionId::new("rev-1"),
);

let outcome = trainer.train_window_once(&experiment)?;
println!("published {}", outcome.head.head_id.as_str());

keep your existing burn model, optimizer, scheduler, and loaders.

burn_p2p handles:

head sync
window-by-window training publication
checkpoint/artifact movement
validator promotion flow
assigned shard fetch for distributed runs

one validator/authority node must exist somewhere on the network to initialize and promote heads.

data

trainers fetch only their assigned shard: with_sharded_dataset(...).

browser peers fetch assigned shards from an http dataset origin, not over the p2p overlay.

what the repo includes

burn_p2p: core runtime
burn_p2p_browser: browser runtime
burn_p2p_bootstrap: bootstrap/browser-edge http surface
burn_p2p_app: reference dioxus ui

same experiment layout works across native and browser peers. browser-facing runtime and ui live in the companion crates above.

see it working

single-machine mixed-fleet mnist sanity run:

cargo xtask e2e mnist --profile smoke --keep-artifacts

best follow-up docs: