optirs_core/distributed/
mod.rs

1// Distributed optimization support
2//
3// This module provides support for distributed training including parameter averaging,
4// gradient compression, and communication optimization for multi-node/multi-GPU training.
5
6pub mod fedprox;
7pub use fedprox::*;
8
9pub mod ring_allreduce;
10pub use ring_allreduce::{CollectiveTransport, LocalTransport, ReduceOp, RingAllReduce};
11
12pub mod pipeline_parallel;
13pub use pipeline_parallel::{
14    OpKind, PipelineConfig, PipelineExecution, PipelineMetrics, PipelineOp, PipelineSchedule,
15    PipelineScheduler, StageCost, StagePartitioner, StageRange,
16};
17
18pub mod elastic;
19pub use elastic::{
20    ElasticConfig, ElasticCoordinator, EpochSnapshot, MembershipEvent, RendezvousState, ShardRange,
21    WorkerShards, WorkerState,
22};
23
24mod averaging;
25pub use averaging::{AveragingStrategy, ParameterAverager};
26
27mod parameter_server;
28pub use parameter_server::ParameterServer;
29
30mod coordinator;
31pub use coordinator::{CommunicationResult, DistributedCoordinator, TrainingStats};
32
33mod compression;
34pub use compression::{
35    CompressedGradient, CompressionMetadata, CompressionStats, CompressionStrategy,
36    GradientCompressor,
37};
38
39#[cfg(test)]
40mod tests;
optirs_core/distributed/mod.rs

optirs_core/distributed/
mod.rs
