Research map / Large language models / Agents, code and math
SWERank: Software Issue Localization with Code Ranking
Revanth Gangi Reddy, Tarun Suresh, JaeHyeok Doo, Ye Liu, Xuan-Phi Nguyen, Yingbo Zhou, Semih Yavuz, Caiming Xiong, Heng Ji, Shafiq Joty
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| apply top-K consistency filtering (K=20) to SweLoc training data instead of no filtering (ablation) | SweRankEmbed-Small · SWE-Bench-Lite | localization accuracy | Improved |
| increase training data proportion (even 5%) of filtered SweLoc (ablation) | SweRankEmbed-Small · SWE-Bench-Lite | localization accuracy | Improved |
| finetune CodeRankEmbed on SweLoc | CodeRankEmbed · SWE-Bench-Lite | Func. Acc@10 (%): 59.5 -> 72.3 | Improved |
| finetune Arctic-Embed on SweLoc | Arctic-Embed · SWE-Bench-Lite | Func. Acc@10 (%): 53.7 -> 71.9 | Improved |
| finetune Qwen-2.5-Text 32B reranker on SweLoc | Qwen-2.5-Text (32B) · SWE-Bench-Lite | Func. Acc@10 (%): 82.5 -> 86.1 | Improved |
| finetune Qwen-2.5-Text 7B reranker on SweLoc | Qwen-2.5-Text (7B) · SWE-Bench-Lite | Func. Acc@5 (%): 75.2 -> 75.6 | No clear effect |
| add SweRankLLM reranking on top of SweRankEmbed-Large (ablation) | SweRankEmbed-Large · SWE-Bench-Lite function | Acc@10 (%): 82.12 -> 88.69 | Improved |
| use SweRank localization instead of LocAgent for downstream repair | SWE-Fixer · SWE-Bench-Lite | Repair Pass@1 (%): 22.6 -> 24.5 | Improved |
| add second iteration of hard-negative mining vs first iteration (ablation) | SweRankEmbed · SWE-Bench-Lite | localization accuracy | Improved |
| reduce embedding dimension from 1024 to 64 (Matryoshka) (ablation) | Qwen3-Embedding-0.6B · SWE-Bench-Lite | Acc@10 (%): 75.18 -> 60.22 | Hurt |
| scale encoder from SweRankEmbed-Large (7B) to Qwen3-Embedding-8B (deeper) (ablation) | retriever · SWE-Bench-Lite | Acc@10 (%): 82.12 vs 83.94 | No clear effect |
| add SweLoc finetuning to RankZephyr-finetuned Llama-3.1 8B reranker (ablation) | Llama-3.1 8B Instruct reranker · SWE-Bench-Lite | Acc@10 (%): 80.29 -> 85.77 | Improved |
Closest papers by meaning
- OrcaLoca: An LLM Agent Framework for Software Issue Localization (ICML 2025 · similarity 0.89)
- CoRNStack: High-Quality Contrastive Data for Better Code Retrieval and Reranking (ICLR 2025 · similarity 0.88)
- Multi-SWE-bench: A Multilingual Benchmark for Issue Resolving (NeurIPS 2025 · similarity 0.88)
- Improving Code Localization with Repository Memory (ICLR 2026 · similarity 0.86)
- Nemotron-CORTEXA: Enhancing LLM Agents for Software Engineering Tasks via Improved Localization and Solution Diversity (ICML 2025 · similarity 0.85)
- SWT-Bench: Testing and Validating Real-World Bug-Fixes with Code Agents (NeurIPS 2024 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
