docs: add KEP-2839 Dynamic LLM Trainer Framework proposal

[NarayanaSabari]

Summary

Add KEP-2839: Dynamic LLM Trainer Framework proposal.

This KEP proposes decoupling the BuiltinTrainer from TorchTune by introducing a pluggable
LLMBackend interface in the Python SDK and a corresponding LLMBackendStrategy in the Go
control plane. TorchTune becomes the first backend implementation (preserving backward
compatibility), and TRL is added as the first new backend with SFT/DPO support.

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

Tracking issue: #2839

What This KEP Covers

SDK (Python)

• LLMBackend abstract base class with to_command(), to_args(), framework(), validate()
• Backend registry with @register_backend decorator supporting external/out-of-tree backends
• BuiltinTrainer.config type widened from TorchTuneConfig to LLMBackend
• TorchTuneConfig refactored to implement LLMBackend (zero breaking changes)
• TRLConfig backend with SFT and DPO trainer types

Go Control Plane

• LLMBackendStrategy interface replacing hardcoded TorchTune command-sniffing in the Torch plugin
• TorchTuneStrategy (wraps existing torchtune.go logic unchanged)
• TRLStrategy (minimal -- TRL config is fully constructed by the SDK)
• Dispatch via trainer.kubeflow.org/framework label on ClusterTrainingRuntime

Infrastructure

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

Non-Goals

• Unsloth or LlamaFactory backends (future work)
• CRD schema changes
• New Kubernetes resource topologies

Test Plan

• Unit tests for backend registry, TorchTuneConfig backward compat, TRLConfig, Go strategy dispatch
• Integration tests for TRL TrainJob reconciliation and runtime listing
• E2E tests for TRL SFT/DPO on GPU and TorchTune regression

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

[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! 🙏