Research map / Large language models / Language, safety and interpretability
QiMeng-CRUX: Narrowing the Gap Between Natural Language and Verilog via Core Refined Understanding eXpression
Lei Huang, Rui Zhang, Jiaming Guo, Yang Zhang, Di Huang, Shuyao Cheng, Pengwei Jin, Chongxiao Li, Zidong Du, Xing Hu, Yunji Chen, Qi Guo
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 |
|---|---|---|---|
| train on RealSpec (degraded realistic descriptions) (ablation) | CodeV SFT 7B · VerilogEval-v2 Spec-to-RTL (T=0) | pass@1: 44.8 -> 53.2 | Improved |
| train on RealSpec (ablation) | CodeV SFT 7B · VerilogEval-v1 Machine | pass@10: 90.1 -> 89.8 | No clear effect |
| add CRUX as SFT target (ablation) | CodeV + RealSpec · RTLLM-v2 | pass@1: 46.2 -> 52.3 | Improved |
| add CRUX as SFT target (ablation) | CodeV + RealSpec · VerilogEval-v2 Spec-to-RTL (T=0.8) | pass@1: 47.2 -> 57.0 | Improved |
| add CRUX as SFT target (ablation) | CodeV + RealSpec · VerilogEval-v1 Machine | pass@1: 79.2 -> 76.4 | Hurt |
| RealSpec alone (ablation) | CodeV · RTLLM-v2 | pass@1: 41.0 -> 46.2 | Improved |
| add CRUX-Reward in GRPO (ablation) | QiMeng-CRUX · VerilogEval-v2 Spec-to-RTL (T=0) | pass@1: 62.8 -> 64.7 | Improved |
| add CRUX-Reward in GRPO (ablation) | QiMeng-CRUX · RTLLM-v2 | pass@1: 59.5 -> 63.8 | Improved |
| add CRUX-Reward in GRPO (ablation) | QiMeng-CRUX · VerilogEval-v1 Machine | pass@10: 91.4 -> 90.2 | Hurt |
| add CRUX-Reward in GRPO (ablation) | QiMeng-CRUX · RTLLM-v2 | pass@5: 73.0 -> 70.6 | Hurt |
| add CRUX-Reward in GRPO (ablation) | QiMeng-CRUX · VerilogEval-v1 Human | pass@1: 61.7 -> 65.2 | Improved |
| prompt general code models with CRUX alone or CRUX plus description (ablation) | general code models · Spec-to-RTL | pass@1 | Improved |
Closest papers by meaning
- DeepRTL: Bridging Verilog Understanding and Generation with a Unified Representation Model (ICLR 2025 · similarity 0.92)
- VeriThoughts: Enabling Automated Verilog Code Generation using Reasoning and Formal Verification (NeurIPS 2025 · similarity 0.88)
- QiMeng-CodeV-R1: Reasoning-Enhanced Verilog Generation (NeurIPS 2025 · similarity 0.87)
- CraftRTL: High-quality Synthetic Data Generation for Verilog Code Models with Correct-by-Construction Non-Textual Representations and Targeted Code Repair (ICLR 2025 · similarity 0.87)
- BetterV: Controlled Verilog Generation with Discriminative Guidance (ICML 2024 · similarity 0.87)
- QiMeng-SALV: Signal-Aware Learning for Verilog Code Generation (NeurIPS 2025 · similarity 0.86)
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.
