atomr-accel-train 0.1.0

Distributed training blueprints on atomr-accel-cuda: DataParallelTrainer, PipelineParallelTrainer, TensorParallelTrainer, AsyncParameterServer, optimizer + loss enums.

Documentation

Coverage
15.53%
16 out of 103 items documented0 out of 31 items with examples
Size
Source code size: 62.81 kB This is the summed size of all the files inside the crates.io package for this release.
Documentation size: 1.92 MB This is the summed size of all files generated by rustdoc for all configured targets
Ø build duration
this release: 42s Average build duration of successful builds.
all releases: 36s Average build duration of successful builds in releases after 2024-10-23.
Links
Homepage
rustakka/atomr-accel
1 0 5
crates.io
Dependencies
Versions
Owners

atomr-accel-train

Distributed training blueprints on atomr-accel-cuda: data-parallel, pipeline-parallel, tensor-parallel, async parameter server, plus typed OptimizerKind / LossKind enums.

Add to your project

[dependencies]
atomr-accel       = "0.0"
atomr-accel-train = "0.0"

use atomr_accel_train::prelude::*;

This crate depends on atomr-accel-cuda (foundation) plus atomr-accel-patterns (for replica routing). Nothing else.

What's inside

Blueprint	Type
Data parallel	`DataParallelTrainer<P>`
Pipeline parallel	`PipelineParallelTrainer<P>`
Tensor parallel	`TensorParallelTrainer<P>`
Async parameter server	`AsyncParameterServer`
Optimizer enum	`OptimizerKind` (SGD, AdamW)
Loss enum	`LossKind` (MSE, CrossEntropy)

License: Apache-2.0.