Top AI Repos — open-source AI, indexed and scored
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Official code for TOIS2026 "Direct Retrieval-augmented Optimization: Synergizing Knowledge Selection and Language Models"
| Date | Stars |
|---|---|
| 2026-07-31 | 266 |
| 2026-08-04 | 266 |
| 2026-08-06 | 266 |
Today
— stars today
This week
— stars this week
This month
— stars this month
Momentum
0.0
growth rate 0.00%/day
<p align="center"> <img src="./assets/background.png" width="100%" alt="Self-Play for Selector-Generator Synergy"> </p> If you like our project, please give us a star ⭐ on GitHub for the latest update. <div align="center"> <h1>Direct Retrieval-augmented Optimization</h1> </div> Retrieval-augmented generation (RAG) is powerful but still limited by a lack of synergy between knowledge selection and answer generation process. Our project introduces a novel framework named Direct Retrieval-augmented Optimization (DRO) that brings Selector and Generator into one collaborative loop, allowing them to learn from each other and optimize jointly. Forget one-sided fine-tuning—we enable mutual enhancement for better factuality, relevance, and performance. If you're working on knowledge-grounded LLMs, this is a drop-in upgrade you don't want to miss. Please checkout our <a href="https://arxiv.org/abs/2505.03075">paper</a> for more details. <div align="center"> <img src="assets/method.png" > </div> ## News - **[2026.1.13]** Our paper has been accepted by the [TOIS 2026](https://dl.acm.org/journal/tois) 🎉🎉🎉! - **[2025.5.05]** The first version of our paper has been released in arxiv. See our paper in this [link](https://arxiv.org/abs/2505.03075). - **[2025.2.01]** Our code was released, including main code for evaluation and training script. ## Environment 1. create the `python` env with `conda` ```txt conda create -n rag python=3.10 conda activate dro pip install -r requirements.txt pip install pytrec_eval -i https://pypi.tuna.tsinghua.edu.cn/simple conda activate dro ``` 2. [Optional] set the `vllm` environment variable when using it. ```txt VLLM_WORKER_MULTIPROC_METHOD=spawn ``` 3. [Optional] Please login the `wandb` if use it to record the loss. ```txt wandb login ``` (`wandb login --relogin` to force relogin) # Dataset ## Download raw dataset | Dataset | Note | Link | | :------------------ |:-----------------------------------------------------------------------------------------------------------------------------------------|:-------------------------------------------------------------------------------------------------------------:| | Nature Question | [Natural Questions: A Benchmark for Question Answering Research](https://aclanthology.org/Q19-1026/) | [Github](https://github.com/facebookresearch/DPR/blob/main/dpr/data/download_data.py)<br/>(click the NQ link) | | HotpotQA | [HotpotQA: A Dataset for Diverse, Explainable Multi-hop Question Answering](https://arxiv.org/abs/1809.09600) | [Homepage](https://hotpotqa.github.io/) | | MuSiQue | [MuSiQue: Multihop Questions via Single-hop Question Composition](https://arxiv.org/abs/2108.00573) | [Github](https://github.com/StonyBrookNLP/musique) | | 2WikiMultihopQA | [Constructing A Multi-hop QA Dataset for Comprehensive Evaluation of Reasoning Steps](https://aclanthology.org/2020.coling-main.580.pdf) | [Github](https://github.com/Alab-NII/2wikimultihop) | | Wizard of Wikipedia | [Wizard of Wikipedia: Knowledge-Powered Conversational agents](https://arxiv.org/abs/1811.01241) | [Google drive](https://drive.google.com/file/d/1z327WOKr70rC4UfOlQVBQnuLxChi_uPs/view?usp=share_link) | ## Retrieval Corpus We follow previous work and use the Wikipedia as our document corpus, which can be found in [DPR](https://github.com/facebookresearch/DPR/blob/main/dpr/data/downlo
Excerpt of 43,103 characters
Read on GitHubWould you bet a product on this? Bounded 0–100 and slow moving.
matched fp:72a3f1b99b2aeef7, topic:retrieval-augmented-generation, desc:retrieval augmented
matched fp:72a3f1b99b2aeef7, topic:large-language-models