Research map / Large language models / Agents, code and math
ParEVO: Synthesizing Code for Irregular Data: High-Performance Parallelism through Agentic Evolution
Liu Yang, Zeyu Nie, Andrew Liu, Ruomu Zou, Deniz Altınbüken, Amir Yazdanbakhsh, Quanquan C. Liu
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 |
|---|---|---|---|
| add evolutionary coding agent with 10 iterations (ablation) | Gemini-3-Pro · DMOJ training problems | speedup: 1.00 -> 1.498 | Improved |
| add evolutionary coding agent with 30 iterations (ablation) | Gemini-3-Pro · DMOJ training problems | speedup: 1.00 -> 2.218 | Improved |
| fine-tune on Parlay-Instruct and DMOJ | Gemini-2.5-Pro · ParEval graph problems | Pass@1: 0.42 -> 0.76 | Improved |
| fine-tune on Parlay-Instruct and DMOJ | Gemini-2.5-Pro · ParEval graph problems | Speedup@1: 21.76 -> 13.67 | Hurt |
| fine-tune on Parlay-Instruct and DMOJ | Gemini-2.5-Pro · ParEval graph problems | Build@1: 0.62 -> 0.97 | Improved |
| fine-tune on Parlay-Instruct and DMOJ | Gemini-2.5-Pro · ParEval C++ ParlayLib | Speedup@1 (AM): 85.34 -> 107.43 | Improved |
| fine-tune on Parlay-Instruct and DMOJ | Gemini-2.5-Pro · ParEval C++ ParlayLib | Pass@1: 0.54 -> 0.58 | Improved |
| fine-tune on ParlayLib syntax and slow-fast pairs | DeepSeek-6.7B-Base · ParEval C++ ParlayLib | Speedup@1 (AM): 0.73 -> 126.16 | Improved |
| fine-tune on ParlayLib syntax and slow-fast pairs | DeepSeek-6.7B-Base · ParEval C++ ParlayLib | Pass@1: 0.11 -> 0.26 | Improved |
| fine-tune with SFT and DPO for Rust | Qwen3-Coder-30B-Instruct · ParEval Rust Rayon | Speedup@1 (AM): 4.57 -> 4.62 | No clear effect |
| fine-tune on Parlay data | Gemini-2.5-Pro · ParEval convex hull | Pass@1: 0.45 -> 0.00 | Hurt |
| fine-tune on Parlay data | Gemini-2.5-Pro · ParEval closest pair 2D | Speedup@1: 74.48 -> 188.03 | Improved |
Closest papers by meaning
- When Parallelism Pays Off: Cohesion-Aware Task Partitioning for Multi-Agent Coding (NeurIPS 2026 · similarity 0.86)
- Towards Direct Latent-Space Synthesis for Parallel Branches in LLM-Agent Workflows (NeurIPS 2026 · similarity 0.85)
- EvoOptiGraph: Weakness-Driven Coevolution via Graph-Based Structural Generation for Optimization Modeling (NeurIPS 2026 · similarity 0.85)
- EvoCodeBench: Evaluating Coding Agents in Multi-Turn Iterative Interactions (NeurIPS 2026 · similarity 0.85)
- Simple Baselines are Competitive with Code Evolution (NeurIPS 2026 · similarity 0.85)
- Improving Parallel Program Performance with LLM Optimizers via Agent-System Interfaces (ICML 2025 · similarity 0.85)
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.
