RTL Generation on RTLLM v2
71.8Pass@1VeriRL (reproduced)
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| VeriRL (reproduced)Role=Fine-tuned, Base Model=Qwen2.5-C-7B-Instruct, SFT=true, RL=true2026.06 | 71.8 | 77.6 | 78.8 | |
| GPT-OSS-120BRole=Teacher, SFT=null, RL=null2026.06 | 69.9 | 78.1 | 80.8 | |
| STGRole=Fine-tuned, Base Model=Qwen3-8B, SFT=true, RL=false2026.06 | 68.7 | 79.9 | 81.9 | |
| CodeV-R1Role=Fine-tuned, Base Model=Qwen2.5-C-7B-Instruct, SFT=true, RL=true2026.06 | 68 | 77.6 | 80.7 | |
| SiliconMind-V1Role=Fine-tuned, Base Model=Qwen3-4B-Thinking, SFT=true, RL=false2026.06 | 67.9 | 75.3 | 76 | |
| STGRole=Fine-tuned, Base Model=Qwen3-4B-Thinking, SFT=true, RL=false2026.06 | 67.5 | 78.2 | 79.8 | |
| SiliconMind-V1Role=Fine-tuned, Base Model=Qwen3-8B, SFT=true, RL=false2026.06 | 66.6 | 74.9 | 76.5 | |
| SiliconMind-V1Role=Fine-tuned, Base Model=Qwen2.5-C-7B-Instruct, SFT=true, RL=false2026.06 | 63.8 | 74 | 75.9 | |
| VeriRL (paper)Role=Fine-tuned, Base Model=Qwen2.5-C-7B-Instruct, SFT=true, RL=true2026.06 | 63.3 | 70.3 | — | |
| STGRole=Fine-tuned, Base Model=Qwen2.5-C-7B-Instruct, SFT=true, RL=false2026.06 | 63.1 | 76.4 | 79 | |
| Qwen3-8BRole=Base, SFT=null, RL=null2026.06 | 40.2 | 61.1 | 67.6 | |
| Qwen3-4B-ThinkingRole=Base, SFT=null, RL=null2026.06 | 36.4 | 50.9 | 56.3 | |
| Qwen2.5-C-7B-InstructRole=Base, SFT=null, RL=null2026.06 | 29.3 | 48.6 | 56 |