HenryCai11/OpenRLHF

An Easy-to-use, Scalable and High-performance RLHF Framework (70B+ PPO Full Tuning & Iterative DPO & LoRA & Mixtral)

Open-source / Comprehensive / Lightweight / Easy-to-use

---

[ English | 中文 ]

OpenRLHF is a high-performance RLHF framework built on Ray, DeepSpeed and HF Transformers:

• Simple and easy to use: OpenRLHF is one of the simplest high-performance RLHF libraries currently available, and seamlessly compatible with Huggingface models and datasets.
• High performance: RLHF training spends 80% of the time on the sample generation stage. Thanks to the ability to use a large inference batch size with Ray and Adam Offload (Pinned Memory) and vLLM generation acceleration, the performance of OpenRLHF 2x+ that of Optimized DeepSpeedChat with Hybrid Engine.
• Distributed RLHF: OpenRLHF distribute the Actor, Reward, Reference, and Critic models onto separate GPUs using Ray, while placing the Adam optimizer on the CPU. This enables full-scale fine-tuning of 70B+ models with multiple A100 80G GPUs and vLLM and 7B models across multiple 24GB RTX 4090 GPUs.
• PPO Implementation Optimization: We integrated the implementation tricks for PPO to improve the training stability, referencing Zhihu and the Notion blog.

More details are in Slides | Technical Report | Documents

Features

• Distributed PPO based on Ray.
• Support full RLHF fine-tuning of models with over 70 billion parameters.
• Support vLLM generation acceleration in RLHF (--vllm_num_engines).
• Support multiple reward models (--reward_pretrain model1,model2...) and remote reward model(--remote_rm_url).
• Support DPO (direct-preference-optimization)/IPO/cDPO.
• Support Kahneman-Tversky optimization (KTO).
• Support Rejection Sampling.
• Support Iterative DPO (https://github.com/RLHFlow/Online-RLHF).
• Support Conditional SFT (https://arxiv.org/abs/2308.12050).
• Support Knowledge Distillation (https://github.com/microsoft/LMOps/tree/main/minillm).
• Support Process Reward Model (PRM).
• Support SFT/DPO/RM/PRM/PPO training samples packing (--packing_samples).
• Support RingAttention (--ring_attn_size, --ring_head_stride)
• Support MoE (--aux_loss_coef)
• Support FlashAttention2 (--flash_attn).
• Support QLoRA (--load_in_4bit), LoRA (--lora_rank, --target_modules).
• Support HuggingFace tokenizer.apply_chat_template in datasets (--apply_chat_template and --input_key).
• Support Wandb log (--use_wandb) and tensorboard (--use_tensorboard).
• Support for recovering from checkpoint (--load_checkpoint and --save_steps).
• Multi-nodes training scripts for Slurm.

PPO Support Matrix

Feature	OpenRLHF	DSChat	CAIChat	TRL
70B+ Full Tuning with 16 A100-80GB	✅	❌	❌	❌
7B Full Tuning with 4 RTX4090	✅	❌	❌	❌
34B DPO Full Tuning with 8 A100-80GB	✅	❌	❌	❌
Inference Engine in PPO	✅	✅	❌	❌
PPO Implementation Tricks	✅	❌	❌	✅
Support QLoRA	✅	❌	❌	✅
Support Mixtral 8*7b	✅	❌	❌	❌
Support Unmerged Actor-Critic	✅	✅	✅	❌
Support Multiple Reward Models	✅	❌	❌	❌
Support Huggingface Models	✅	✅	✅	✅
Easy-to-use	✅	❌ (HybridEngine bugs)	✅	✅

Quick Start

Installation

To use OpenRLHF, first launch the docker container (Recommended) and pip install openrlhf inside the docker container:

# Launch the docker container
docker run --runtime=nvidia -it --rm --shm-size="10g" --cap-add=SYS_ADMIN -v $PWD:/openrlhf nvcr.io/nvidia/pytorch:24.02-py3 bash
sudo pip uninstall xgboost transformer_engine flash_attn -y

# pip install
pip install openrlhf

# If you want to use vLLM acceleration (To install vLLM 0.4.2)
pip install openrlhf[vllm]
# latest vLLM is also supported (Please use `--vllm_sync_backend gloo` or `export NCCL_P2P_DISABLE=1`)
pip install openrlhf[vllm_latest]

# pip install the latest version
pip install git+https://github.com/OpenRLHF/OpenRLHF.git

# Or git clone
git clone https://github.com/OpenRLHF/OpenRLHF.git
cd OpenRLHF
pip install -e .

Note

We recommend using vLLM 0.4.2, as the 0.4.3+ versions currently require synchronizing weights via Gloo (--vllm_sync_backend gloo) or disabling P2P communication (export NCCL_P2P_DISABLE=1). We also provided the Dockerfiles for vLLM and One-Click Installation Script of Nvidia-Docker.

Prepare Datasets

OpenRLHF provides multiple data processing methods in our dataset classes. Such as in the Prompt Dataset:

def preprocess_data(data, input_template=None, input_key="input", apply_chat_template=None) -> str:
    if apply_chat_template:
        prompt = apply_chat_template(data[input_key], tokenize=False, add_generation_prompt=True)
    else:
        prompt = data[input_key]
        if input_template:
            prompt = input_template.format(prompt)
    return prompt

• We can use --input_key to specify the JSON key name of the input datasets --prompt_data {name or path} (PPO) or --dataset {name or path}, and use --apply_chat_template to utilize the chat_template from the Huggingface Tokenizer.
• If you don't want to use --apply_chat_template, you can use --input_template instead, or preprocess the datasets offline in advance.
• OpenRLHF also support mixing multiple datasets using --prompt_data_probs 0.1,0.4,0.5 (PPO) or --dataset_probs 0.1,0.4,0.5.

How Chat Templating Works:

dataset = [{"input_key": [
  {"role": "user", "content": "Hello, how are you?"},
  {"role": "assistant", "content": "I'm doing great. How can I help you today?"},
  {"role": "user", "content": "I'd like to show off how chat templating works!"},
]}]

tokenizer.apply_chat_template(dataset[0]["input_key"], tokenize=False)

"<s>[INST] Hello, how are you? [/INST]I'm doing great. How can I help you today?</s> [INST] I'd like to show off how chat templating works! [/INST]"

How to specify training and test datasets ?

You can specify it using the data_type@data_dir format. For example, the dataset can be set as --dataset json@./data.

data
├── test.jsonl
└── train.jsonl

Note

By default, we use train and test as splits to distinguish training and testing datasets from Huggingface. The JSON key options depends on the specific datasets. See Reward Dataset and SFT Dataset

Supervised Fine-tuning

OpenRLHF's model checkpoint is fully compatible with HuggingFace models. You can specify the model name or path using --pretrain {name or path}, --reward_pretrain {name or path} and --critic_pretrain {name or path}. We have provided some pre-trained checkpoints and datasets on HuggingFace OpenRLHF.

Then you can use the startup scripts we provide in the examples/scripts directory, or start the training using the following commands.

deepspeed --module openrlhf.cli.train_sft \
   --max_len 4096 \
   --dataset Open-Orca/OpenOrca \
   --input_key question \
   --output_key response \
   --input_template 'User: {}\nAssistant: ' \
   --train_batch_size 256 \
   --micro_train_batch_size 2 \
   --max_samples 500000 \
   --pretrain meta-llama/Meta-Llama-3-8B \
   --save_path ./checkpoint/llama3-8b-sft \
   --save_steps -1 \
   --logging_steps 1 \
   --eval_steps -1 \
   --zero_stage 2 \
   --max_epochs 1 \
   --bf16 \
   --flash_attn \
   --learning_rate 5e-6 \
   --gradient_checkpointing \
   --use_wandb {wandb_token}

# Support HF tokenizer.apply_chat_template
# --apply_chat_template 
# --input_key {JSON Key}
# --tokenizer_chat_template {HF Chat Template}

# Support samples packing
# --packing_samples

# Can also be used for continued pre-training
# --pretrain_mode

Note

OpenRLHF SFT/DPO/RewardModel/PPO trainers support --packing_samples based on --flash_attn

Reward Model Training

deepspeed --module openrlhf.cli.train_rm \
   --save_path ./checkpoint/llama3-8b-rm \
   --save_steps -1 \
   --logging_steps 1 \
   --eval_steps -1 \
   --train_batch_size 256 \
   --micro_train_batch_size 1 \
   --pretrain OpenRLHF/Llama-3-8b-sft-mixture \
   --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 \
   --use_wandb {wandb_token}

# Support samples packing
# --packing_samples

PPO without Ray

deepspeed --module openrlhf.cli.train_ppo \
  --pretrain OpenRLHF/Llama-3-8b-sft-mixture \
  --reward_pretrain OpenRLHF/Llama-3-8b-rm-mixture \
  --save_path ./checkpoint/llama-3-8b-rlhf \
  --save_steps -1 \
  --logging_steps 1 \
  --eval_steps -1 \
  --micro_train_batch_size 2 \
  --train_batch_size 128 \
  --micro_rollout_batch_size 4 \
  --rollout_batch_size 1024 \
  --max_epochs 1 \
  --prompt_max_len 1024 \
  --generate_max_len 1024 \
  --zero_stage 2 \
  --bf16 \
  --actor_learning_rate 5e-7 \
  --critic_learning_rate 9e-6 \
  --init_kl_coef 0.01 \
  --prompt_data OpenRLHF/prompt-collection-v0.1 \
  --input_key context_messages \
  --apply_chat_template \
  --max_samples 100000 \
  --normalize_reward \
  --adam_offload \
  --flash_attn \
  --gradient_checkpointing \
  --use_wandb {wandb_token}

# Support remote reward model (HTTP)
# --remote_rm_url http://localhost:5000/get_reward

PPO with Ray and vLLM

To improve RLHF training speed or support 70B models, we can use the PPO with Ray and vLLM acceleration

# launch the master node of ray in container
ray start --head --node-ip-address 0.0.0.0 --num-gpus 8

# if you want to launch ray on more nodes, use
ray start --address {MASTER-NODE-ADDRESS}:6379  --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 2 \
  --reward_num_nodes 1 \
  --reward_num_gpus_per_node 2 \
  --critic_num_nodes 1 \
  --critic_num_gpus_per_node 2 \
  --actor_num_nodes 1 \
  --actor_num_gpus_per_node 2 \
  --vllm_num_engines 2 \
  --vllm_tensor_parallel_size 2 \
  --colocate_critic_reward \
  --colocate_actor_ref \
  --pretrain OpenRLHF/Llama-3-8b-sft-mixture \
  --reward_pretrain OpenRLHF/Llama-3-8b-rm-mixture \
  --save_path /openrlhf/examples/checkpoint/llama3-8b-rlhf \
  --micro_train_batch_size 8 \
  --train_batch_size 128 \
  --micro_rollout_batch_size 16 \
  --rollout_batch_size 1024 \
  --max_samples 100000 \
  --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 \
  --prompt_data OpenRLHF/prompt-collection-v0.1 \
  --input_key context_messages \
  --apply_chat_template \
  --normalize_reward \
  --adam_offload \
  --flash_attn \
  --gradient_checkpointing \
  --use_wandb {wandb_token}

# Support samples packing
# --packing_samples (Recommended)

# Support remote reward model (HTTP)
# --remote_rm_url http://localhost:5000/get_reward

Note

Do not set --vllm_num_engines means not using the vLLM engine. You can also use setup_commands to let Ray automatically deploy the environment, such as --runtime-env-json='{"setup_commands": ["pip install openrlhf[vllm]"]}'.

The launch scripts and documents for supported algorithms are in example/scripts and Documents - Usage

Performance

We optimized DSChat's performance to the greatest extent possible by employing techniques such as enabling Adam offload, along with reward model (RM) and reference model (Ref) offload to increase the micro-batch size during the inference stage and avoid out-of-memory issues. We even fixed some bugs in DSChat to enable the Hybrid Engine (HE) for LLaMA2. The average time (seconds) it took to train 1024 prompts with 1 PPO epoch using the Optimized DSChat and OpenRLHF:

Size	NVIDIA A800-80GB GPUs	Optimized DSChat (with Hybrid Engine)	OpenRLHF	Speedup
7B	16	855.09	471.11	1.82x
13B	32	1528.93	608.93	2.5x
34B	32	3634.98	1526.4	2.4x
70B	32	10407.0	4488.53	2.3x

Note

The data is outdated; please refer to the performance tuning section for re-testing.

Performance Tuning Guide

To achieve optimal performance, we recommend allocating more nodes to the vLLM Engine. For example, for a 70B model with 32 A100 GPUs, it is advised to allocate more than 16 A100 GPUs to the vLLM Engine, 8 GPUs to the Actor model, and the remaining 8 GPUs to the Critic model. Additionally, enable the --colocate_critic_reward, --colocate_actor_ref, or --ref_reward_offload (Optional) options to merge nodes. Finally, you should increase the rollout_micro_batch_size (and minimize the TP size of vLLM engine) as much as possible, and avoid OOM (Out Of Memory) using --packing_samples. During the training phase, a larger --micro_train_batch_size is better. Enable enable_prefix_caching in vLLM generation when n_samples_per_prompt > 1.

Companies and Organizations using OpenRLHF

• ByteDance
• NexusFlow
• Baidu
• Jülich Supercomputing Centre (JSC)
• Berkeley Starling Team
• Tencent
• Alibaba
• Google
• China Telecom
• ...

Join Us

How to Join?

1. Email us at janhu9527@gmail.com or join GitHub Organization. Please include the following details:
   • Your name
   • Your GitHub username
   • Your areas of interest
   • Your skills and experience related to NLP and/or AI
2. You can also join us through the official GitHub OpenRLHF ↗ project page. Just create an issue about your interest to contribute and we will get back to you.

What can you do?

1. Join the team and participate in the development of the OpenRLHF project.
2. Contribute to the project by submitting pull requests.
3. Help improve documentation, fix bugs, or create new features.
4. Share the project and help us grow the community.

Sponsor Us

Your sponsorship can help us maintain and improve OpenRLHF. If you find this project useful, please consider sponsoring us. You can sponsor us on Open Collective ↗.

Starchart

Contributors

A big thank you to all our contributors! If you want to contribute, feel free to make a pull request or create an issue.

References & Acknowledgements

We would like to express our gratitude to the following projects and organizations for their contributions to the field of AI and NLP:

• Hugging Face Transformers ↗
• OpenAI GPT ↗
• LLaMA ↗
• DeepSpeed ↗
• Ray ↗

Our project would also like to thank ColossalChat and DeepSpeedChat. In the early stages of the project, we referred to their code design.

(2024/7) Our GitHub organization has changed from OpenLLMAI to OpenRLHF.

Citation

@article{hu2024openrlhf,
  title={OpenRLHF: An Easy-to-use, Scalable and High-performance RLHF Framework},
  author={Jian Hu and Xibin Wu and Weixun Wang and Xianyu and Dehao Zhang and Yu Cao},
  journal={arXiv preprint arXiv:2405.11143},
  year={2024}
}

---

OpenRLHF © 2024 OpenRLHF. All Rights Reserved.