Where is the InfoSeek-Evaluation set?

[nv-aken12]

Hi, thank you for your great work!!
I read InfoSeek and InfoFlow papers.

In the second paper (InfoFlow), I found a paragraph that states InfoSeek has an Evaluation set containing 300 queries:

InfoSeek-Evaluation The InfoSeek-Evaluation set contains 300 high-quality, human-checked
samples to evaluate agentic deep search capability. Qwen2.5-72B-Instruct with a CoT prompting
achieves lower than 8% accuracy in this evaluation set.

However, I could not find this evaluation set in the InfoSeek repository or its HuggingFace repository.
Could you please let me know where I can access the InfoSeek-Evaluation set?

Thank you for your time, and I look forward to your response.

References:

• OPEN DATA SYNTHESIS FOR DEEP RESEARCH
• INFOFLOW: REINFORCING SEARCH AGENT VIA REWARD DENSITY OPTIMIZATION

[kunluo531]

Hi, thank you for your interest in our work!

The InfoSeek evaluation set is available in this HF repository:
https://huggingface.co/datasets/Lk123/InfoSeek

We also recently released a more challenging benchmark for scientific-paper deep research, AutoResearchBench:

• Paper: https://arxiv.org/pdf/2604.25256
• Dataset: https://huggingface.co/datasets/Lk123/AutoResearchBench

We recommend using AutoResearchBench for evaluating more advanced deep research capabilities. Thank you!