AGENT SKILL

openrlhf-training

High-performance RLHF framework with Ray+vLLM acceleration. Use for PPO, GRPO, RLOO, DPO training of large models (7B-70B+). Built on Ray, vLLM, ZeRO-3. 2× faster than DeepSpeedChat with distributed architecture and GPU resource sharing.

License 许可
MIT
最近更新
2026-10-22
Stars
★ 1
Forks
0

01 前言 Front Matter

name
openrlhf-training
tags
[ "Post-Training", "OpenRLHF", "RLHF", "PPO", "GRPO", "RLOO", "DPO", "Ray", "vLLM", "Distributed Training", "Large Models", "ZeRO-3" ]
author
Orchestra Research
license
MIT
version
1.0.0
description
High-performance RLHF framework with Ray+vLLM acceleration. Use for PPO, GRPO, RLOO, DPO training of large models (7B-70B+). Built on Ray, vLLM, ZeRO-3. 2× faster than DeepSpeedChat with distributed architecture and GPU resource sharing.
dependencies
[ "openrlhf", "ray", "vllm", "torch", "transformers", "deepspeed" ]

02 资源文件 4

03 SKILL.md 正文

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

04 同仓库更多 Skills

CZX-LBJ
CZX-LBJ/deepseek-science

bio-free-energy-calculations

Performs alchemical free-energy calculations including relative binding free energy (RBFE / FEP+) and absolute binding free energy (ABFE) via OpenFE, FEP+, GROMACS, AMBER pmemd, and OpenMM with explicit lambda scheduling, soft-core potentials, MBAR/BAR analysis, cycle-closure validation, and protocol-appropriate enhanced sampling. Compares ML alternatives (Boltz-2 affinity, DeepDock). Use when ranking analogs by binding affinity beyond docking accuracy, performing prospective lead optimization, or validating SAR predictions.

★ 1
CZX-LBJ
CZX-LBJ/deepseek-science

bio-conformer-generation

Generates 3D conformer ensembles using RDKit ETKDGv3 with knowledge-enhanced distance geometry, MMFF94/UFF force-field optimization, CREST + GFN2-xTB semi-empirical refinement, and macrocycle-aware torsion preferences. Provides explicit decision rules for single vs ensemble conformer use, RMSD pruning, energy windows, conformer count, and force-field choice. Use when preparing 3D ligands for docking, generating descriptor input for 3D QSAR, or sampling macrocycle/peptide conformational ensembles.

★ 1
CZX-LBJ
CZX-LBJ/deepseek-science

bio-admet-prediction

Predicts ADMET properties using ADMETlab 3.0 (119 platform features, including 77 prediction models with modeled-endpoint uncertainty), ADMET-AI, DeepChem MolNet, and chemprop D-MPNN with explicit handling of OECD QSAR principles, applicability domain assessment, calibration, hERG/CYP/AMES endpoints, and PAINS / Lipinski / Ro5 / Veber / BBB druglikeness filters. Use when filtering compounds for drug-likeness, prioritizing leads by predicted safety, or building an in-house ADMET QSAR model.

★ 1
CZX-LBJ
CZX-LBJ/deepseek-science

writing-systems-papers

Paragraph-level structural blueprint for 10-12 page systems papers targeting OSDI, SOSP, ASPLOS, NSDI, and EuroSys. Provides page allocation, paragraph templates, and writing patterns. Use when user says "写系统论文", "systems paper structure", "OSDI paper", "SOSP paper", or wants fine-grained structural guidance for a systems conference submission.

★ 1
CZX-LBJ
CZX-LBJ/deepseek-science

wiki-enrich

Fill in the per-paper TODO sections of research-wiki/papers/<slug>.md pages that literature-ingest skills leave as bare scaffolds. Use when user says 'enrich wiki', 'fill paper TODOs', 'wiki body 補完', '把 paper 摘要寫進 wiki', 'research-wiki 自動填', or after a batch ingest that left papers/ as TODO scaffolds.

★ 1
CZX-LBJ
CZX-LBJ/deepseek-science

web-debug-search

Search GitHub, Stack Exchange, Chinese technical communities, official documentation, and general developer web sources for software errors, compatibility problems, API usage questions, and real-world workarounds. Use for debugging and discovery only; results are not paper-citation evidence.

★ 1