@inproceedings{bajpai-etal-2026-exploring,
    title = "Exploring Capability Thresholds in Ultra-Lightweight {LLM} Judges for Nugget-Based Report Evaluation",
    author = "Bajpai, Mann  and
      Chatwal, Pulkit  and
      Deswal, Priyanshu  and
      Singh, Harish Pratap  and
      Mishra, Santosh Kumar",
    editor = "Yang, Eugene  and
      Lawrie, Dawn  and
      MacAvaney, Sean  and
      Mayfield, James  and
      Soldaini, Luca  and
      Yates, Andrew",
    booktitle = "Proceedings of the 1st Workshop on Multilingual Report Generation via Retrieval Augmented Generation ({RAG}4{R}eports 2026)",
    month = jul,
    year = "2026",
    address = "San Diego, CA, USA",
    publisher = "Association for Computational Linguistics",
    url = "https://preview.aclanthology.org/ingest-acl-workshops/2026.rag4reports-1.13/",
    pages = "94--98",
    ISBN = "979-8-89176-417-0",
    abstract = "Reliable automatic evaluation of retrieval-grounded long-form reports typically requires human annotation or frontier-scale proprietary LLMs, both of which are expensive in constrained settings. Team rgipt participated in RAG4Reports@ACL 2026 Task 1 with a zero-shot nugget-verification system that runs entirely on a single NVIDIA T4 GPU. We compare three ultra-lightweight decoder-only models: Qwen2-0.5B, Qwen2-1.5B, and Qwen2.5-0.5B, under identical inference conditions to examine how small an LLM judge can be while retaining human-aligned ranking signal. Both Qwen2 models produced negative $\tau_{\text{gap}}$, whereas Qwen2.5-0.5B achieved $\tau_{\text{gap}} = 0.0772$ and Pearson $r = 0.2209$, ranking 13th of 21 teams. Within this family and evaluation setting, model generation appears to matter more than parameter count, although this finding is based on three configurations on a single task and warrants further validation."
}