# run-house/kubetorch

Distribute and run AI workloads on Kubernetes magically in Python, like PyTorch for ML infra.

Repository: https://github.com/run-house/kubetorch
Canonical: https://ross.abutalabs.com/products/kubetorch
Homepage: https://run.house
Language: Python
License: Apache-2.0
License Family: permissive
Topics: artificial-intelligence, aws, data-science, gcp, machine-learning, python, pytorch, ray, serverless, distributed, infrastructure, observability, data-processing, evaluation, inference, kubernetes, training
Last push: 2026-05-29T13:37:59+00:00

## Health v2 (maintenance only)
Score: 83/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 84, release rhythm 71, longevity 100
- inputs: {"age_days": 1576, "days_push": 96, "days_rel": 197, "gap_med": 7.5, "n_releases_24m": 23}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1224, forks 60 (observed 2026-08-28T04:04:02.794508+00:00)

## What it is
Kubetorch is a Python library that lets you distribute and run ML workloads (training, inference, data processing) on Kubernetes directly from Python code, without local runtimes or code serialization. It provides a serverless-like interface with fast iteration, real-time log/error propagation, and built-in fault handling, deployed via a Helm chart.

## Use cases
- run distributed ML training on kubernetes from python
- deploy model inference on a k8s cluster
- iterate on RL training jobs quickly with remote GPUs
- run python functions on remote cluster compute
- scale ML workloads with bin-packing and autoscaling
- replace ray cluster setup with simpler python API
- run ML evaluation jobs on kubernetes

## When to choose
- you want a Pythonic API to run ML workloads on Kubernetes without writing YAML
- you need fast iteration loops for distributed training or RL
- you want built-in fault recovery and resource adjustment for ML jobs
- you already have a Kubernetes cluster and want serverless-like ML compute

## When to avoid
- you don't use Kubernetes and just need local or single-machine training
- you need a fully managed platform without operating a cluster
- you only need simple batch scheduling without ML-specific features
- you require a very mature, battle-tested orchestrator like Ray or Kubeflow

## Facets
- artifact type: library
- maturity: active
- function: machine-learning, llm-training, llm-inference, deployment, container-orchestration, serverless, workflow-automation, sdk
- domain: machine-learning, deep-learning, artificial-intelligence, data-science, cloud-computing, infrastructure-as-code, microservices
- platform: python, cloud, self-hosted
- tags: distributed-training, ml-infrastructure, remote-execution, gpu-computing, helm-chart, pytorch, ray-alternative, devops, kubernetes, docker

## Member repositories
- run-house/kubetorch (main) score 83

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:04:02.794508+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T06:15:29.438889+00:00, confidence not recorded.
  - readme: https://github.com/run-house/kubetorch (fetched 2026-08-28T04:04:02.794508+00:00, sha bfa4b4a2fd9f)
  - homepage: https://run.house (fetched 2026-08-29T12:23:56.940122+00:00, sha f0bcfa5ebeab)
  - registry_pypi: https://pypi.org/pypi/kubetorch/json (fetched 2026-08-29T12:23:56.950219+00:00, sha dceefbb7914c)
- Data as of 2026-08-30T08:39:29.467469+00:00.