| Paper | Model | Dataset | Data Pipeline | Training | Evaluation |

• 🌐 4-round data pipeline — turns raw Wikipedia seeds into a verified multi-hop QA dataset through generation, self-verification, and external judging
• 💸 Zero API spend — Zero dollars spent on expensive Web Search APIs or search-enabled LLMs; the entire pipeline runs on Selenium + DeepSeek Chat and uses search aggregators such as DDGS
• 🤖 GRPO training with web search — trains search-augmented LLMs via verl-tool with a parallel multi-backend DDGS retrieval server
• 📊 12-benchmark evaluation suite — covers single-hop (NQ, TriviaQA, PopQA), multi-hop (HotpotQA, 2WikiMHQA, MuSiQue), and hard reasoning (FRAMES, GAIA, MoNaCo) benchmarks

• 🗂️ Data Pipeline
• 🌱 Round 1 — Seed Creation
• ❓ Round 2 — QA Generation
• ✅ Round 3 — Self-Verification
• 🏅 Round 4 — External Verification
• 🏋️ Model Training
• 📐 Data Processing
• 🔎 Evaluation

ORBIT QA pairs are built through four sequential rounds. See data/README.md for the full overview.

Installing uv itself:

curl -LsSf https://astral.sh/uv/install.sh | sh

uv sync
source .venv/bin/activate

Prior to training & evaluating search agents, you should prepare the training, eval and test datasets. See models/data_process for the full data processing scripts.

export HF_HOME=/u3/n3thakur/projects/cache
export HF_TOKEN=hf_...

Training data — ORBIT + NQ + HotpotQA mixed at a 1:1:1 ratio:

python models/data_process/prepare_train_data.py \
    --datasets 'nq,hotpotqa,orbit-ai/orbit-20k' \
    --local_dir models/train/data/mix-nq-hotpotqa-orbit-ratio-1-1-1 \
    --ratio 1:1:1

Eval data — all 12 validation benchmarks:

python models/data_process/prepare_eval_data.py \
    --data_sources nq,triviaqa,popqa,hotpotqa,2wikimultihopqa,musique,bamboogle,frames,gaia,monaco,webwalkerqa,webshaper \
    --local_dir models/eval/data/all-12-val-datasets

Test data — 8 Wikipedia test benchmarks:

python models/data_process/prepare_test_data.py \
    --local_dir models/eval/data/all-8-wikipedia-test-datasets

To train search agents using ORBIT, please see models/train.

export HF_TOKEN="hf_..."
export WANDB_API_KEY="..."

bash models/train/ddgs_web_search.sh # run ddgs web server
bash models/train/run_grpo.sh # train search agent with verl-tool

To evaluate search agents trained using ORBIT, please see models/eval.

bash models/eval/retrieval_server_bge.sh # run BGE index using faiss (conda env required)
bash models/eval/run_eval.sh # evaluate search agent on Wikipedia datasets

orbit/
├── data/
│   ├── round-1-seed-creation/         # Wikipedia pages → seed JSONL
│   ├── round-2-qa-generation/         # Seeds → inverted QA pairs (DeepSeek)
│   ├── round-3-self-verification/     # QA pairs → self-verified answers (DeepSeek)
│   ├── round-4-external-verification/ # Verified pairs → externally judged (vLLM)
│   └── outputs/                       # Intermediate + final JSONL files
├── models/
│   ├── train/                         # GRPO training + DDGS retrieval server
│   ├── eval/                          # BGE retrieval index + evaluation
│   └── data_process/                  # Prepare train/eval parquet files
├── pyproject.toml
└── README.md

We thank the following open-source projects:

• verl-tool for the GRPO training framework for training search agents.
• vLLM for LLM inference.
• DDGS for a web search aggregator tool (used for search agent training with GRPO).
• Search-R1 for retrieval server design inspiration and initial experiments.

If you have any questions or suggestions, please contact us at:

• Nandan Thakur: [email protected]
• Zijian Chen: [email protected]
• Xueguang Ma: [email protected]

If you find this data generation repository helpful, please cite our preprint work ORBIT: Scalable and Verifiable Data Generation for Search Agents on a Tight Budget:

@misc{thakur2026orbit,
      title={ORBIT: Scalable and Verifiable Data Generation for Search Agents on a Tight Budget}, 
      author={Nandan Thakur and Zijian Chen and Xueguang Ma and Jimmy Lin},
      year={2026},
      eprint={2604.01195},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2604.01195}, 
}