Kubeflow Trainer

https://www.kubeflow.org/docs/components/trainer/overview/

• Designed for llm fine-tuning
• Enabling scalable, distributed training
• Support various frameworks(torch, jax, tensorflow)

You can develop your LLMs with:

1. Python SDK
2. K8s Custom Resources API -> Kubernetes Training Runtimes

Optimize GPU utilization and gang-scheduling for ML workloads by leveraging Kubernetes projects like

• Kueue
• Coscheduling
• Volcano
• YuniKorn

Docs

V1::doc

V2::doc

V2::md