docs: KEP-2839 Dynamic LLM Trainer Framework — ConfigTrainer for BaseTrainer hierarchy

[NarayanaSabari]

Summary

Redesigned KEP-2839: Dynamic LLM Trainer Framework proposal.

This KEP introduces a pluggable config-driven trainer framework for LLM fine-tuning, aligned with KEP-285 (Specialized Trainer Abstractions).

What This KEP Proposes

SDK (Python)

• ConfigTrainer base class within KEP-285's BaseTrainer hierarchy for config-driven trainers
• TorchTuneTrainer(ConfigTrainer) — refactored from TorchTuneConfig with backward-compatible alias
• TRLTrainer(ConfigTrainer) — new backend with SFT/DPO/KTO/GRPO support
• Runtime auto-discovery via trainer.kubeflow.org/framework label
• BuiltinTrainer.config type widened from TorchTuneConfig to ConfigTrainer

Go Control Plane

• FrameworkStrategy interface replacing hardcoded TorchTune command-sniffing in the Torch plugin
• TorchTuneStrategy (wraps existing logic unchanged)
• TRLStrategy (accelerate-compatible env var injection)
• Label-based dispatch via trainer.kubeflow.org/framework

Infrastructure

• TRL container image (cmd/trainers/trl/)
• TRL ClusterTrainingRuntime manifests
• Helm chart additions for TRL runtimes

Relationship to KEP-285

This KEP answers the open question from the KEP-285 review about how config-driven trainers fit into the BaseTrainer hierarchy:

                    BaseTrainer (ABC)                ← KEP-285
                         │
         ┌───────────────┼───────────────┐
         │               │               │
   TorchTrainer     JAXTrainer     ConfigTrainer      ← This KEP
   (func-based)     (func-based)   (config-driven)
                                        │
                         ┌──────────────┼──────────────┐
                         │              │              │
                  TorchTuneTrainer  TRLTrainer    (future backends)

Non-Goals

• Unsloth or LlamaFactory backends (future work)
• CRD schema changes
• Deprecating BuiltinTrainer or CustomTrainer

Test Plan

• Unit tests for ConfigTrainer interface, backend registry, TRL arg generation, Go strategy dispatch
• Integration tests for TRL TrainJob reconciliation
• E2E tests for TRL SFT/DPO on GPU and TorchTune regression

Builds on KEP-2401 and the community consensus on "Plan 3" from #2752.

Tracking issue: #2839

/cc @Electronic-Waste @andreyvelich @tariq-hasan @szaher

[google-oss-prow]

[APPROVALNOTIFIER] This PR is NOT APPROVED

This pull-request has been approved by:
Once this PR has been reviewed and has the lgtm label, please assign terrytangyuan for approval. For more information see the Kubernetes Code Review Process.

The full list of commands accepted by this bot can be found here.

Details

Needs approval from an approver in each of these files:

• OWNERS

Approvers can indicate their approval by writing /approve in a comment
Approvers can cancel approval by writing /approve cancel in a comment

[github-actions]

🎉 Welcome to the Kubeflow Trainer! 🎉

Thanks for opening your first PR! We're happy to have you as part of our community 🚀

Here's what happens next:

• If you haven't already, please check out our Contributing Guide for repo-specific guidelines and the Kubeflow Contributor Guide for general community standards.
• Our team will review your PR soon! cc @kubeflow/kubeflow-trainer-team

Join the community:

• Slack: Join our #kubeflow-trainer Slack channel.
• Meetings: Attend the Kubeflow AutoML and Training Working Group bi-weekly meetings.

Feel free to ask questions in the comments if you need any help or clarification!
Thanks again for contributing to Kubeflow! 🙏