Balancing Reasoning and Hardware Constraints in RAG Pipelines for Ukrainian Multi-Domain Document Understanding

A resource-efficient Hybrid RAG pipeline using BM25, BGE-M3, and Cross-Encoder reranking scored 0.8095 on the private test set of the UNLP 2026 Shared Task on Multi-Domain Document Understanding, placing 10th out of 15 active teams. The team reported that OCR of scanned Ukrainian PDFs consumed 5-7 hours of the 9-hour offline Kaggle execution limit under sequential single-threaded execution, leaving roughly two hours of LLM inference for 500 questions, so they ran a 4-bit quantized LapaLLM 12B model via llama.cpp on dual NVIDIA T4 GPUs instead of parameter-heavy reasoning models such as DeepSeek R1, which consistently timed out.

arXiv:2609.22124v1 Announce Type: new Abstract: This paper describes the system submitted to the UNLP 2026 Shared Task on Multi-Domain Document Understanding. The challenge required extracting precise answers, document IDs, and page numbers from a diverse corpus of Ukrainian PDF documents within a strict 9-hour offline Kaggle execution limit. During evaluation on the hidden private test set, optical character recognition OCR of scanned documents emerged as a severe bottleneck, consuming 5-7 hours of the total time budget due to sequential single-threaded execution. This overhead strictly limited the remaining time for Large Language Model LLM inference to approximately two hours for 500 questions. To guarantee pipeline completion without timeouts, we developed a resource-efficient Hybrid Retrieval-Augmented Generation RAG pipeline utilizing BM25, BGE-M3, and Cross-Encoder reranking. Rather than deploying parameter-heavy reasoning models e.g., DeepSeek R1 which consistently timed out, we utilized a 4-bit quantized LapaLLM 12B model via llama.cpp on dual NVIDIA T4 GPUs. Prioritizing pipeline stability over multi-step reasoning, our system achieved a Private Score of 0.8095, placing 10th out of 15 active teams.