# kai-scheduler/KAI-Scheduler

KAI Scheduler is an open source Kubernetes Native scheduler for AI workloads at large scale

Repository: https://github.com/kai-scheduler/KAI-Scheduler
Canonical: https://ross.abutalabs.com/products/kai-scheduler
Language: Go
License: Apache-2.0
License Family: permissive
Last push: 2026-08-26T18:59:01+00:00

## Health v2 (maintenance only)
Score: 82/100 (v2, computed 2026-09-03T02:39:23.370411+00:00)
- activity 99, release rhythm 86, longevity 39
- inputs: {"age_days": 553, "days_push": 7, "days_rel": 14, "gap_med": 1, "n_releases_24m": 150}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1472, forks 253 (observed 2026-08-28T04:04:49.401961+00:00)

## What it is
KAI Scheduler is a Kubernetes-native scheduler written in Go that optimizes GPU resource allocation for AI and machine learning workloads at large scale. It supports the full AI lifecycle from small interactive jobs to large training and inference workloads, with fairness, gang scheduling, and topology-aware placement across clusters of thousands of nodes.

## Use cases
- schedule GPU workloads on a large Kubernetes cluster
- run training and inference jobs fairly across teams
- gang scheduling for distributed AI training jobs
- topology-aware placement for disaggregated LLM serving
- maximize GPU utilization in a shared AI cluster
- dynamically allocate GPUs to pods in Kubernetes

## When to choose
- you operate a large multi-tenant GPU cluster on Kubernetes
- you need fair-share GPU allocation between teams
- you run both interactive notebooks and large distributed training in one cluster
- you need topology-aware or gang scheduling for AI workloads

## When to avoid
- you have a small cluster with no GPU workloads
- the default kube-scheduler already meets your needs
- you need scheduling for non-Kubernetes environments

## Facets
- artifact type: infra-config
- maturity: active
- function: container-orchestration, scheduling, gpu-computing, deployment
- domain: infrastructure-as-code, machine-learning, cloud-computing, gpu-computing
- platform: go, cloud, self-hosted
- tags: kubernetes-scheduler, gpu-scheduling, gang-scheduling, ai-workloads, fair-share, topology-aware-scheduling, training-jobs, inference, containers, kubernetes, linux

## Member repositories
- kai-scheduler/KAI-Scheduler (main) score 82

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:04:49.401961+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 2026-08-30T04:34:41.961378+00:00, confidence not recorded.
  - readme: https://github.com/kai-scheduler/KAI-Scheduler (fetched 2026-08-28T04:04:49.401961+00:00, sha 9b8e04917dfc)
- Data as of 2026-08-30T08:39:29.467469+00:00.