Single-controller LoRA RL fine-tuning with vLLM #735

gursimar · 2025-12-16T23:42:45Z

Description

This PR adds working, tested examples for running single-controller LoRA training with the vLLM backend.
It builds on the existing LoRA + vLLM support (RFC #609) and demonstrates how to configure and launch a single-controller GRPO workflow.

What’s included

New GRPO example for single-controller LoRA with vLLM
Corresponding YAML config illustrating minimal wiring
No changes to core engine or runtime behavior

Files changed

Kept files in the examples/lora folder on purpose as IMO, all lora exmaples should be under this forder only.

examples/lora/gsm8k_grpo_vllm_single_controller.py — single-controller GRPO LoRA example
examples/lora/gsm8k_grpo_vllm_single_controller.yaml — config for vLLM backend

Running instructions

python examples/lora/gsm8k_grpo_vllm_single_controller.py --config examples/lora/gsm8k_grpo_vllm_single_controller.yaml

Testing

Verified the example runs end-to-end with the vLLM backend.
No regressions to existing LoRA or non-LoRA flows.

Type of Change

Checklist

I have read the Contributing Guide
I have run formatting tools (pre-commit or manual)
I have run relevant unit tests and they pass
I have added tests for new functionality
I have updated documentation if needed
My branch is up to date with main
This PR introduces breaking changes
No critical issues raised by AI reviewers (/gemini review)

Need help? Check the Contributing Guide or ask in
GitHub Discussions!

gemini-code-assist · 2025-12-16T23:43:02Z

Summary of Changes

Hello @gursimar, I'm Gemini Code Assist¹! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed!

This pull request delivers a practical and verified example for fine-tuning models using LoRA with a single controller and the vLLM inference engine. It serves as a blueprint for users looking to implement GRPO workflows with these specific technologies, providing both the Python script and the necessary YAML configuration to get started. The primary goal is to expand the existing LoRA + vLLM capabilities with a concrete, runnable demonstration.

Highlights

New Example for LoRA Fine-tuning: This pull request introduces a new, fully tested example demonstrating single-controller LoRA (Low-Rank Adaptation) training utilizing the vLLM backend.
GRPO Workflow Configuration: It provides a clear example of how to configure and launch a single-controller GRPO (Generalized Reinforcement Learning with Policy Optimization) workflow, building on existing LoRA + vLLM support.
Dedicated Configuration File: A corresponding YAML configuration file is included, illustrating the minimal wiring required to set up this specific LoRA fine-tuning scenario.
No Core Engine Changes: The changes are confined to examples and configuration, ensuring no modifications to the core engine or runtime behavior of the system.

Using Gemini Code Assist

The full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips.

Invoking Gemini

You can request assistance from Gemini at any point by creating a comment using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands on the current page.

Feature	Command	Description
Code Review	`/gemini review`	Performs a code review for the current pull request in its current state.
Pull Request Summary	`/gemini summary`	Provides a summary of the current pull request in its current state.
Comment	@gemini-code-assist	Responds in comments when explicitly tagged, both in pull request comments and review comments.
Help	`/gemini help`	Displays a list of available commands.

Customization

To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a .gemini/ folder in the base of the repository. Detailed instructions can be found here.

Limitations & Feedback

Gemini Code Assist may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments. If you're interested in giving your feedback about your experience with Gemini Code Assist for Github and other Google products, sign up here.

You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension.

Review the Privacy Notices, Generative AI Prohibited Use Policy, Terms of Service, and learn how to configure Gemini Code Assist in GitHub here. Gemini can make mistakes, so double check it and use code with caution. ↩

gemini-code-assist

Code Review

This pull request introduces a new example for single-controller LoRA fine-tuning with the vLLM backend. The changes include a Python script for the training workflow and a corresponding YAML configuration file. The code is well-structured for an example script. My review includes a couple of suggestions for the Python script to improve maintainability by removing a magic number and to add a placeholder for an evaluation step, which seems intended by the configuration but is currently missing.

examples/lora/gsm8k_grpo_vllm_single_controller.py

garrett4wade

While the implementation looks great, I'd like to still confirm the details about learning performance.

The previous SPMD LoRA code has an unresolved bug that if multiple infernece engines submit rollout requests concurrently, the learning performance will significantly drop. As a workaround, we only submit requests on rank 0 (code). Only through this way the learning curve can basically match full-parameter tuning.

I wonder whether the bug still exists in the single controller mode. Can you provide learning curves comparing this new script with the default SPMD, full-parameter tuning script? Hopefully there is no performance drop any more.

garrett4wade · 2025-12-17T09:47:28Z

examples/lora/gsm8k_grpo_vllm_single_controller.py

+            with stats_tracker.record_timing("save"):
+                saver.save(actor, epoch, step, global_step, tokenizer=tokenizer)
+
+            with stats_tracker.record_timing("checkpoint_for_recover"):
+                recover_handler.dump(
+                    actor,
+                    step_info,
+                    saver,
+                    evaluator,
+                    stats_logger,
+                    train_dataloader,
+                    tokenizer=tokenizer,
+                )


The single-controller training script has been slightly changed. It now has an additional clear_batch call. Please refer to the latest script for details.

FYI we are working to merging the scripts into trainers now.

garrett4wade · 2025-12-17T09:48:10Z

examples/lora/gsm8k_grpo_vllm_single_controller.py

+                if config.rollout.max_head_offpolicyness > 0:
+                    batch = actor.prepare_batch(
+                        train_dataloader,
+                        workflow="areal.workflow.rlvr.RLVRWorkflow",
+                        workflow_kwargs=workflow_kwargs,
+                    )
+                else:
+                    batch = actor.rollout_batch(
+                        next(data_generator),
+                        workflow="areal.workflow.rlvr.RLVRWorkflow",
+                        workflow_kwargs=workflow_kwargs,
+                    )


We should only use prepare_batch. Please check the latest script in the main branch.

… vllm backend

gemini-code-assist bot reviewed Dec 16, 2025

View reviewed changes

examples/lora/gsm8k_grpo_vllm_single_controller.py Show resolved Hide resolved

examples/lora/gsm8k_grpo_vllm_single_controller.py Show resolved Hide resolved

gursimar force-pushed the single_controller_lora branch from 12d9725 to fd4fd16 Compare December 17, 2025 00:19

garrett4wade reviewed Dec 17, 2025

View reviewed changes

gursimar added 2 commits December 17, 2025 15:03

Working and tested examples for grpo single controller lora using the…

d7b2ed6

… vllm backend

Cleaned up some debug statements

90b7da1

gursimar force-pushed the single_controller_lora branch from fd4fd16 to 90b7da1 Compare December 17, 2025 23:03

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Single-controller LoRA RL fine-tuning with vLLM #735

Single-controller LoRA RL fine-tuning with vLLM #735

Uh oh!

gursimar commented Dec 16, 2025

Uh oh!

gemini-code-assist bot commented Dec 16, 2025

Uh oh!

gemini-code-assist bot left a comment

Uh oh!

Uh oh!

Uh oh!

garrett4wade left a comment

Uh oh!

garrett4wade Dec 17, 2025

Uh oh!

garrett4wade Dec 17, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Single-controller LoRA RL fine-tuning with vLLM #735

Are you sure you want to change the base?

Single-controller LoRA RL fine-tuning with vLLM #735

Uh oh!

Conversation

gursimar commented Dec 16, 2025

Description

What’s included

Files changed

Running instructions

Testing

Type of Change

Checklist

Uh oh!

gemini-code-assist bot commented Dec 16, 2025

Summary of Changes

Highlights

Footnotes

Uh oh!

gemini-code-assist bot left a comment

Choose a reason for hiding this comment

Code Review

Uh oh!

Uh oh!

Uh oh!

garrett4wade left a comment

Choose a reason for hiding this comment

Uh oh!

garrett4wade Dec 17, 2025

Choose a reason for hiding this comment

Uh oh!

garrett4wade Dec 17, 2025

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants