Analog Circuit Sizing on Analog Circuit Sizing (ACS) Out-of-domain (unseen tasks)
0.14OPAMP 1 ErrorQwen3-30B-Instruct-2507
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Qwen3-30B-Instruct-2507Eval=Multi-turn2026.01 | 0.14 | 0.01 | 0.05 | 0.16 | 0.09 | |
| Qwen3-30B-Instruct-2507Eval=ST-iter2026.01 | 0.14 | 0.04 | 0.06 | 0.15 | 0.1 | |
| Trajectory-Level GRPO (base: Qwen3-30B-Instruct-2507)Eval=Multi-turn2026.01 | 0.15 | 0.18 | 0.2 | 0.17 | 0.18 | |
| DeepSeek-V3.2 (thinking)Eval=ST-iter2026.01 | 0.2 | 0.28 | 0.32 | 0.23 | 0.26 | |
| Single-Turn GRPO (base: Qwen3-30B-Instruct-2507)Eval=ST-iter2026.01 | 0.2 | 0.15 | 0.26 | 0.25 | 0.21 | |
| Turn-Level GRPO (base: Qwen3-30B-Instruct-2507)Eval=Multi-turn2026.01 | 0.2 | 0.32 | 0.31 | 0.58 | 0.35 | |
| Qwen3-235B-Instruct-2507Eval=Multi-turn2026.01 | 0.21 | 0.04 | 0.08 | 0.19 | 0.13 | |
| DeepSeek-V3.2 (no-think)Eval=ST-iter2026.01 | 0.22 | 0.26 | 0.24 | 0.24 | 0.24 | |
| DeepSeek-V3.2 (thinking)Eval=Multi-turn2026.01 | 0.22 | 0.23 | 0.26 | 0.26 | 0.24 | |
| Qwen3-235B-Instruct-2507Eval=ST-iter2026.01 | 0.23 | 0.17 | 0.15 | 0.24 | 0.2 | |
| DeepSeek-V3.2 (no-think)Eval=Multi-turn2026.01 | 0.25 | 0.23 | 0.24 | 0.3 | 0.26 | |
| Bayesian OptimizationEval=null2026.01 | 0.36 | 0.01 | 0.08 | 0.24 | 0.17 |