Streamline LLM Fine-tuning on Kubernetes With Kubeflow LLM Trainer
summary
Kubeflow LLM Trainer provides a Kubernetes-native fine-tuning pipeline that hides YAML and Docker chores behind TrainingRuntime CRD and lightweight TrainJob overrides, exposes a Python SDK to start work with a one-line launch, and automates model and dataset fetch plus artifact export. Demo shows Llama 3.2 family and Q2.5 models, LoRA adapter export, per-node resource overrides (examples used 35 GB), and jobs that run to completion, proving the blueprints plus controller approach actually works.
AI-written summary · about
chapters (7)
- 0:00Introduction
- 1:47Fine-tuning challenges
- 3:55Custom resource solution
- 6:44Flexible fine-tuning
- 8:30Streamlined training workflow
- 10:39Live demo: Kubeflow LLM Trainer
- 15:03Future roadmap
related talks
21mStreamline LLM Fine-Tuning on Kubernetes with Kubeflow LLM Trainer
25mFine-Tuning Large Language Models With Declarative ML Orchestration
20mEmpowering ML Workloads With Kubeflow: JAX Distributed Training and LL...
32mRunning and Fine-tuning Open Source LLMs on Google Kubernetes Engine
51mMachine Learning using Kubeflow and Kubernetes
54mScaling Machine Learning Pipelines in Cloud
29mML Engineering Kubeflow
50mRunning Open LLM Models on Kubernetes