Information Retrieval / 2026

ListT5 Reranker Extension

An inference-time exploration of listwise reranking, grouping strategy, tournament flow, and compute-aware retrieval experiments.

Listwise rerankingBEIR evaluationTournament inferenceCandidate caching

01 / Context

This project explores how ListT5-style listwise reranking behaves under compute-aware inference strategies.

02 / Role

Describe the implementation, experiment design, debugging, and result analysis responsibilities here.

03 / System

Use this section to explain BM25 candidate retrieval, grouping, first-round caching, tournament reranking, and final output selection.

04 / Technical Depth

Good future details include duplicate-index handling, forward-call budgets, dataset differences, and quality-latency trade-offs.

05 / Product / Research Angle

Frame this as a search-quality versus compute-efficiency study, not just a leaderboard exercise.

06 / Result

Add the final table, best configurations, and a short discussion of where the method helped or hurt.

07 / Reflection

Explain what you would test next if there were more compute or time.