OpenRLHF is a high-performance framework for training large models using Reinforcement Learning from Human Feedback (RLHF), optimized for distributed training with Ray and vLLM.

What models can I train with OpenRLHF?

OpenRLHF supports training large models ranging from 7B to over 70B parameters, making it suitable for a variety of advanced machine learning applications.

Which algorithms does OpenRLHF support?

OpenRLHF supports several RLHF training algorithms including PPO, GRPO, RLOO, and DPO, allowing users to choose the best method for their specific use case.

How do I install OpenRLHF?

To install OpenRLHF, you need to run a Docker container with the NVIDIA PyTorch image and then use pip to install the OpenRLHF package along with vLLM support.

What are the hardware requirements for using OpenRLHF?

OpenRLHF requires NVIDIA A100 or H100 GPUs, with specific VRAM requirements depending on the model size. A multi-node Ray cluster setup is recommended for optimal performance.

OpenRLHF Training

Name: OpenRLHF Training
Availability: InStock
Author: davila7

Free

Accelerate RLHF training for large models with Ray and vLLM.

by davila730.2k stars on davila7/claude-code-templates

4 views

Updated Aug 10, 2026

Get this skill

Free · Opens the source repo

What OpenRLHF Training does

OpenRLHF is a high-performance framework designed for Reinforcement Learning from Human Feedback (RLHF) training, leveraging the capabilities of Ray and vLLM for optimized distributed training. This framework supports various algorithms including Proximal Policy Optimization (PPO), Group Normalized Policy Optimization (GRPO), Reinforcement Learning with Off-Policy Optimization (RLOO), and Direct Preference Optimization (DPO), making it versatile for different training needs. With its architecture, OpenRLHF can handle large models ranging from 7B to over 70B parameters, ensuring efficiency and speed in training processes.

The installation process is straightforward, requiring Docker and specific package installations to set up the environment. Once installed, users can initiate training jobs using Ray, which allows for effective resource management across multiple GPUs. The framework is designed to be memory-efficient, particularly with the GRPO algorithm, which does not require a critic model, thus conserving GPU resources while maintaining training performance.

OpenRLHF is particularly beneficial for researchers and developers working on large-scale AI models who need a robust solution for RLHF training. The framework's focus on distributed architecture allows it to scale effectively, making it suitable for teams with access to multi-node GPU clusters. Additionally, the integration of vLLM inference acceleration enhances the training speed, making it up to twice as fast as some existing solutions like DeepSpeedChat.

Overall, OpenRLHF provides a comprehensive solution for those looking to implement advanced RLHF training techniques in their machine learning projects, offering flexibility in algorithm selection and efficient resource utilization.

When to use it

Use OpenRLHF when you need to train large models (7B-70B+) using reinforcement learning techniques and require efficient distributed training capabilities.

When not to use it

Avoid OpenRLHF if you're working with smaller models or if you need a simpler, single-node training solution, as alternatives like TRL may be more suitable in those cases.

What you can build with it

Training a Large Language Model

Use OpenRLHF to train a 70B parameter language model efficiently using PPO and distributed resources.

Implementing Custom Reward Models

Leverage OpenRLHF's capabilities to create and train custom reward models for specific applications in RLHF.

Scaling Training Across Multiple Nodes

Utilize OpenRLHF's distributed architecture to scale your training jobs across a multi-node GPU cluster, maximizing resource utilization.

How to install OpenRLHF Training

View source

1. Install with the skills CLI

npx skills add davila7/claude-code-templates/post-training-openrlhf --agent claude-code

2. Or install it manually

Download the skill folder and drop it into ~/.claude/skills/ for all projects, or .claude/skills/ to scope it to one repo. Restart Claude Code so it picks up the new skill.

Anthropic's agentic coding CLI, and the reference implementation of Agent Skills. Drop a skill folder into ~/.claude/skills and Claude Code loads it automatically whenever a task matches the skill's description. Claude Code docs

Inside SKILL.md

Written by davila7

OpenRLHF - High-Performance RLHF Training

Quick start

OpenRLHF is a Ray-based RLHF framework optimized for distributed training with vLLM inference acceleration.

Installation:

# Launch Docker container
docker run --runtime=nvidia -it --rm --shm-size="10g" --cap-add=SYS_ADMIN \
  -v $PWD:/openrlhf nvcr.io/nvidia/pytorch:25.02-py3 bash

# Uninstall conflicts
sudo pip uninstall xgboost transformer_engine flash_attn pynvml -y

# Install OpenRLHF with vLLM
pip install openrlhf[vllm]

PPO Training (Hybrid Engine):

ray start --head --node-ip-address 0.0.0.0 --num-gpus 8

ray job submit --address="http://127.0.0.1:8265" \
  --runtime-env-json='{"working_dir": "/openrlhf"}' \
  -- python3 -m openrlhf.cli.train_ppo_ray \
  --ref_num_nodes 1 --ref_num_gpus_per_node 8 \
  --reward_num_nodes 1 --reward_num_gpus_per_node 8 \
  --critic_num_nodes 1 --critic_num_gpus_per_node 8 \
  --actor_num_nodes 1 --actor_num_gpus_per_node 8 \
  --vllm_num_engines 4 --vllm_tensor_parallel_size 2 \
  --colocate_all_models \
  --vllm_gpu_memory_utilization 0.5 \
  --pretrain OpenRLHF/Llama-3-8b-sft-mixture \
  --reward_pretrain OpenRLHF/Llama-3-8b-rm-700k \
  --save_path ./output/llama3-8b-rlhf \
  --micro_train_batch_size 8 --train_batch_size 128 \
  --micro_rollout_batch_size 16 --rollout_batch_size 1024 \
  --max_epochs 1 --prompt_max_len 1024 --generate_max_len 1024 \
  --zero_stage 3 --bf16 \
  --actor_learning_rate 5e-7 --critic_learning_rate 9e-6 \
  --init_kl_coef 0.01 --normalize_reward \
  --gradient_checkpointing --packing_samples \
  --vllm_enable_sleep --deepspeed_enable_sleep

GRPO Training (Group Normalized Policy Optimization):

# Same command as PPO, but add:
--advantage_estimator group_norm

Common workflows

Workflow 1: Full RLHF pipeline (SFT → Reward Model → PPO)

Step 1: Train reward model (DPO):

deepspeed --module openrlhf.cli.train_rm \
  --save_path ./output/llama3-8b-rm \
  --save_steps -1 --logging_steps 1 \
  --eval_steps -1 --train_batch_size 256 \
  --micro_train_batch_size 1 --pretrain meta-llama/Meta-Llama-3-8B \
  --bf16 --max_epochs 1 --max_len 8192 \
  --zero_stage 3 --learning_rate 9e-6 \
  --dataset OpenRLHF/preference_dataset_mixture2_and_safe_pku \
  --apply_chat_template --chosen_key chosen \
  --rejected_key rejected --flash_attn --gradient_checkpointing

Step 2: PPO training:

ray start --head --node-ip-address 0.0.0.0 --num-gpus 8

ray job submit --address="http://127.0.0.1:8265" \
  -- python3 -m openrlhf.cli.train_ppo_ray \
  --ref_num_nodes 1 --ref_num_gpus_per_node 8 \
  --reward_num_nodes 1 --reward_num_gpus_per_node 8 \
  --critic_num_nodes 1 --critic_num_gpus_per_node 8 \
  --actor_num_nodes 1 --actor_num_gpus_per_node 8 \
  --vllm_num_engines 4 --vllm_tensor_parallel_size 2 \
  --colocate_all_models \
  --pretrain OpenRLHF/Llama-3-8b-sft-mixture \
  --reward_pretrain ./output/llama3-8b-rm \
  --save_path ./output/llama3-8b-ppo \
  --micro_train_batch_size 8 --train_batch_size 128 \
  --micro_rollout_batch_size 16 --rollout_batch_size 1024 \
  --max_epochs 1 --prompt_max_len 1024 --generate_max_len 1024 \
  --zero_stage 3 --bf16 \
  --actor_learning_rate 5e-7 --critic_learning_rate 9e-6 \
  --init_kl_coef 0.01 --normalize_reward \
  --vllm_enable_sleep --deepspeed_enable_sleep

Workflow 2: GRPO training (no critic model needed)

Memory-efficient alternative to PPO:

ray job submit --address="http://127.0.0.1:8265" \
  -- python3 -m openrlhf.cli.train_ppo_ray \
  --advantage_estimator group_norm \
  --ref_num_nodes 1 --ref_num_gpus_per_node 8 \
  --reward_num_nodes 1 --reward_num_gpus_per_node 8 \
  --actor_num_nodes 1 --actor_num_gpus_per_node 8 \
  --vllm_num_engines 4 --vllm_tensor_parallel_size 2 \
  --colocate_all_models \
  --pretrain OpenRLHF/Llama-3-8b-sft-mixture \
  --reward_pretrain OpenRLHF/Llama-3-8b-rm-700k \
  --save_path ./output/llama3-8b-grpo \
  --micro_train_batch_size 8 --train_batch_size 128 \
  --micro_rollout_batch_size 16 --rollout_batch_size 1024 \
  --max_epochs 1 --bf16 \
  --actor_learning_rate 5e-7 \
  --init_kl_coef 0.01 --use_kl_loss --kl_estimator k3 \
  --normalize_reward --no_advantage_std_norm

Key GRPO parameters:

--advantage_estimator group_norm - Enables GRPO
--use_kl_loss - KL loss from GRPO paper
--kl_estimator k3 - Loss function (k2 ≈ k1)
--no_advantage_std_norm - Disables std normalization

Workflow 3: DPO training (preference optimization)

Simpler alternative without reward model:

deepspeed --module openrlhf.cli.train_dpo \
  --save_path ./output/llama3-8b-dpo \
  --save_steps -1 --logging_steps 1 \
  --eval_steps -1 --train_batch_size 256 \
  --micro_train_batch_size 2 --pretrain meta-llama/Meta-Llama-3-8B \
  --bf16 --max_epochs 1 --max_len 8192 \
  --zero_stage 3 --learning_rate 5e-7 --beta 0.1 \
  --dataset OpenRLHF/preference_dataset_mixture2_and_safe_pku \
  --apply_chat_template --chosen_key chosen \
  --rejected_key rejected --flash_attn --gradient_checkpointing

When to use vs alternatives

Use OpenRLHF when:

Training large models (7B-70B+) with RL
Need vLLM inference acceleration
Want distributed architecture with Ray
Have multi-node GPU cluster
Need PPO/GRPO/RLOO/DPO in one framework

Algorithm selection:

PPO: Maximum control, best for complex rewards
GRPO: Memory-efficient, no critic needed
RLOO: Modified PPO with per-token KL
REINFORCE++: More stable than GRPO, faster than PPO
DPO: Simplest, no reward model needed

Use alternatives instead:

TRL: Single-node training, simpler API
veRL: ByteDance's framework for 671B models
DeepSpeedChat: Integrated with DeepSpeed ecosystem

Common issues

Issue: GPU OOM with large models

Disable model colocation:

# Remove --colocate_all_models flag
# Allocate separate GPUs for each model
--actor_num_gpus_per_node 8 \
--critic_num_gpus_per_node 8 \
--reward_num_gpus_per_node 8 \
--ref_num_gpus_per_node 8

Issue: DeepSpeed GPU index out of range

Set environment variable:

export RAY_EXPERIMENTAL_NOSET_CUDA_VISIBLE_DEVICES=1

Issue: Training instability

Use Hybrid Engine instead of async:

--colocate_all_models \
--vllm_enable_sleep \
--deepspeed_enable_sleep

Adjust KL coefficient:

--init_kl_coef 0.05  # Increase from 0.01

Issue: Slow generation during PPO

Enable vLLM acceleration:

--vllm_num_engines 4 \
--vllm_tensor_parallel_size 2 \
--vllm_gpu_memory_utilization 0.5

Advanced topics

Hybrid Engine GPU sharing: See references/hybrid-engine.md for vLLM sleep mode, DeepSpeed sleep mode, and optimal node allocation.

Algorithm comparison: See references/algorithm-comparison.md for PPO vs GRPO vs RLOO vs REINFORCE++ benchmarks and hyperparameters.

Multi-node setup: See references/multi-node-training.md for Ray cluster configuration and fault tolerance.

Custom reward functions: See references/custom-rewards.md for reinforced fine-tuning and agent RLHF.

Hardware requirements

GPU: NVIDIA A100/H100 recommended
VRAM:
- 7B model: 8× A100 40GB (Hybrid Engine)
- 70B model: 48× A100 80GB (vLLM:Actor:Critic = 1:1:1)
Multi-node: Ray cluster with InfiniBand recommended
Docker: NVIDIA PyTorch container 25.02+

Performance:

2× faster than DeepSpeedChat
vLLM inference acceleration
Hybrid Engine minimizes GPU idle time

Resources

Docs: https://github.com/OpenRLHF/OpenRLHF
Paper: https://arxiv.org/abs/2405.11143
Examples: https://github.com/OpenRLHF/OpenRLHF/tree/main/examples
Discord: Community support

Frequently asked questions about OpenRLHF Training

Similar skills

Spring Boot Testing

Master testing techniques for Spring Boot 4 applications.

GitHub Issues

Manage GitHub issues efficiently with MCP tools.

Geofeed Tuner

Optimize your IP geolocation feeds in CSV format.

Batch Files

Master Windows batch scripting for automation and task management.

Adobe Illustrator Scripting

Automate your Illustrator workflows with ExtendScript.

Plugin Structure

Create and organize Claude Code plugins effectively.