You can deploy this model and start earning money today!

RAG-Gym/
Direct-HotpotQA-DPO

Name: Direct-HotpotQA-DPO
Availability: InStock

peft

safetensors

1910.09700

meta-llama/Llama-3.1-8B-Instruct

adapter

Views

No views yet

Direct-HotpotQA-DPO

This is the DPO actor for the Direct agent on HotpotQA, released as part of RAG-Gym. Direct is a baseline agent used for comparison in RAG-Gym. It is a single-step agent that answers directly without intermediate reasoning or retrieval. This checkpoint was trained with direct preference optimization on process-level preference pairs.

Base model: meta-llama/Meta-Llama-3.1-8B-Instruct
Agent architecture: Direct (a baseline agent used for comparison in RAG-Gym)
Task / dataset: HotpotQA (multi-hop question answering over Wikipedia)
Training method: direct preference optimization (DPO)
Adapter: PEFT LoRA (r=256, alpha=512), task type CAUSAL_LM
Precision: bf16

Intended use

Act as the reasoning/search policy for the Direct agent on HotpotQA-style (multi-hop question answering over Wikipedia) tasks within RAG-Gym.
Can be run with zero-shot inference or, together with the matching PRM critic, with critic-guided (Best-of-N) inference.

This model is intended for research on process-supervised retrieval-augmented generation. It is not intended for clinical decision-making or other high-stakes use.

How to use

Load the LoRA adapter on top of the base model:

python

import torch
from transformers import AutoTokenizer, AutoModelForCausalLM
from peft import PeftModel

base = "meta-llama/Meta-Llama-3.1-8B-Instruct"
tokenizer = AutoTokenizer.from_pretrained(base)
model = AutoModelForCausalLM.from_pretrained(base, torch_dtype=torch.bfloat16)
model = PeftModel.from_pretrained(model, "RAG-Gym/Direct-HotpotQA-DPO")

For full usage within the agent framework (environment setup, inference, and critic-guided Best-of-N selection), see the RAG-Gym repository.

Training

Fine-tuned from meta-llama/Meta-Llama-3.1-8B-Instruct on HotpotQA process-reward data using direct preference optimization (DPO) with the TRL library. Adaptation uses LoRA (rank 256, alpha 512, task type CAUSAL_LM) in bf16. The training code and full configuration are available in the RAG-Gym repository.

Limitations

The model is trained on HotpotQA-style data and is intended for research use within RAG-Gym. Outputs may be inaccurate or unsupported by retrieved evidence and should not be relied upon for high-stakes decisions. As an 8B-parameter model, it requires a CUDA-enabled GPU to run efficiently.

License

Public Domain Notice (U.S. Government Work, NCBI). See the LICENSE in the repository. Please cite the authors when using this material.

Citation

bibtex

@article{xiong2025raggym,
    title={RAG-Gym: Optimizing Reasoning and Search Agents with Process Supervision},
    author={Guangzhi Xiong and Qiao Jin and Xiao Wang and Yin Fang and Haolin Liu and Yifan Yang and Fangyuan Chen and Zhixing Song and Dengyu Wang and Minjia Zhang and Zhiyong Lu and Aidong Zhang},
    journal={arXiv preprint arXiv:2502.13957},
    year={2025}
}