kubeflow/trainer

KEP-2839: Kubeflow Dynamic LLM Trainer Framework

Open

#2,839 opened on Sep 19, 2025

View on GitHub
 (39 comments) (5 reactions) (1 assignee)Go (964 forks)github user discovery
area/llmgood first issuehelp wantedkind/feature

Repository metrics

Stars
 (2,112 stars)
PR merge metrics
 (PR metrics pending)

Description

What you would like to be added?

I would like to add Kubeflow Dynamic LLM Trainer Framework to:

  1. Streamline LLM Fine-tuning on Kubernetes with a simple yet efficient Python SDK
  2. Allow dynamic and extensible registration of popular post-training frameworks
  3. Make it possible for day0&day1 model support

Google Doc: http://bit.ly/4nyrQS8 (work in progress)

Upstream KEP: https://github.com/kubeflow/trainer/issues/2401

Appreciate your feedbacks! @kubeflow/kubeflow-trainer-team @kubeflow/kubeflow-sdk-team @deepanker13 @saileshd1402 @seanlaii @helenxie-bit @varshaprasad96 @franciscojavierarceo @thesuperzapper @rimolive @juliusvonkohout @jbottum @varodrig @Doris-xm

Why is this needed?

Coming from https://github.com/kubeflow/trainer/issues/2752, we decided to adopt the plan 3 after several rounds of discussions.

Love this feature?

Give it a 👍 We prioritize the features with most 👍

Contributor guide

Research direction
Read the KEP document and existing Kubeflow Trainer architecture. Understand the components and workflow for LLM fine tuning.
Tech stack
pythongokubernetes
Domain
aimachine learningbackend
Issue type
Feature
DifficultyEstimated implementation difficulty for a new contributor, from 1 for very small changes to 5 for expert-level work.
3
Estimated timeA rough time range for an experienced contributor to investigate, implement, test, and prepare a pull request.
Over 1 week
Activity statusHow available the issue appears right now: fresh, active, stale, blocked, or waiting on maintainer input.
Active
ClarityHow clearly the issue explains the expected change, acceptance criteria, and next step.
Mostly clear
Prerequisites
PythonKubernetesGit
Newbie friendlinessA 1-100 score estimating how approachable this issue is for first-time contributors.
25