Skip to main content

Coded Compute

A Python toolkit for simulating distributed coded computation, modeling worker stragglers, and running configurable experiments with NumPy and Hydra.

Source code and documentation

Introduction

Coded computation adds redundancy to worker inputs so that outputs can be recovered when some workers do not respond. Coded Compute models this process in four stages:

  1. Generate a code matrix whose rows specify each worker's input coefficients.
  2. Encode the inputs using those coefficients.
  3. Execute user-defined worker operations and collect responding workers' results.
  4. Decode those results using the corresponding code rows.

Workers execute locally and sequentially. Stragglers are modeled as missing responses rather than elapsed-time delays or network failures.

Use cases

  • Prototype code families and study recovery under worker failures.
  • Compare numerical recovery across code matrices and worker calculation bit depth.
  • Run reproducible experiments with seeded probabilistic straggler models.
  • Define experiment setup and execution Hydra using YAML files.

Features

  • Extensible deterministic and randomized code-family interfaces.
  • Coefficient-based input encoding and worker assignment.
  • Per-worker operations, calculation dtypes, and operation arguments.
  • Deterministic and probabilistic straggler wrappers.
  • Decoding with surviving-code rank checks.
  • Configurable pipelines with shared workspace values, loops, and nested pipelines.

Recovery example

With two input arrays, A and B, three workers can compute A + B, A + 2B, and A + 3B. Any two responses are sufficient to reconstruct both arrays.

The repository includes a YAML recovery experiment and its Python components. The experiment configures three workers, marks one as a straggler, decodes the remaining responses, and checks recovery against the original arrays.

Recovery requires enough independent surviving code rows and an operation that preserves the code's linear relationships. Arbitrary nonlinear operations are not generally recoverable through pseudoinverse decoding. Numerical conditioning also affects reconstruction accuracy.

Versioning

Latest Version is v0.5.0 released on Nov 14th, 2026

Installation and usage

Python 3.12 or newer is required.

python -m pip install codedcompute

The Python import name is codedcompute. Runtime dependencies are NumPy, Hydra Core, and OmegaConf.

Start with the core API documentation to define a code family, encode inputs, configure workers, and decode results. For YAML-based experiments, see Hydra pipelines.

License

Licensed under the MIT License. Copyright (c) 2026 Royee Yosibash.

Metadata

Release files for codedcompute 0.5.0

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for codedcompute 0.5.0
File Size Uploaded Provenance Actions
codedcompute-0.5.0.tar.gz 17.3 kB Without provenance Details for codedcompute-0.5.0.tar.gz

Built distribution (wheel)

Table of built distributions (wheels) for codedcompute 0.5.0
File Interpreter ABI Platform Actions
codedcompute-0.5.0-py3-none-any.whl Python 3 none any Details for codedcompute-0.5.0-py3-none-any.whl

Total release size: 37.1 kB

Release files / codedcompute-0.5.0.tar.gz

Download URL codedcompute-0.5.0.tar.gz
Size 17.3 kB
Tags Source
SHA-256 checksum
How to use checksums
c164cae94e9ee88cdc1753014d6bbce2c0b95a1af66a1912fd1ba8a83903d9aa
BLAKE2b-256 checksum
How to use checksums
2241c7b6fef5f3521d8d9886a06d307e8f19a08996a026182682979b7c9347ef
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
No
Uploaded via twine/7.0.0 CPython/3.12.3

Release files / codedcompute-0.5.0-py3-none-any.whl

Download URL codedcompute-0.5.0-py3-none-any.whl
Size 19.8 kB
Tags Python 3
SHA-256 checksum
How to use checksums
dd8c6257f15880c9732b9a1b0ead192c9d9ba9947000bd1c15ce8744bf9f5aee
BLAKE2b-256 checksum
How to use checksums
d96dc699bd12b587ffcab40546dcb6f7f324fc065853f34fdb5f82283348eec7
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
No
Uploaded via twine/7.0.0 CPython/3.12.3

Security for codedcompute 0.5.0

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

No provenance found

None of the files in this release were uploaded with provenance metadata.

A note on package security

PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.

To learn more about protecting yourself from malicious actors, visit our package security documentation.

Release history Release notifications | RSS feed

This release

0.5.0 This release

2 release files No provenance

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page