Deep search agent checkpoint from Qwen2.5-3B-Instruct under
Adaptive Information Control for Search-Augmented LLM Reasoning (EMNLP 2026).
1vllm serve sxiong/DeepControl-Qwen2.5-3B \
2 --dtype bfloat16 \
3 --max-model-len 8192
Loading the model alone does not provide retrieval. It requires the DeepControl system prompt, action parser, and wiki18 corpus.
1@article{xiong2026adaptive,
2 title={Adaptive Information Control for Search-Augmented LLM Reasoning},
3 author={Xiong, Siheng and Gungordu, Oguzhan and Kerce, James C and Fekri, Faramarz},
4 journal={arXiv preprint arXiv:2602.01672},
5 year={2026}
6}