STEM Knowledge on GPQA
31.9Pass@1 AccuracyTAC
Evaluation Results
| Method | Links | |
|---|---|---|
| TACBackbone=Qwen3-1.7B2026.06 | 31.9 | |
| SECBackbone=Qwen3-1.7B2026.06 | 30.7 | |
| RandomBackbone=Qwen3-1.7B2026.06 | 30 | |
| TACBackbone=Llama3.2-3B2026.06 | 28.7 | |
| RandomBackbone=Llama3.2-3B2026.06 | 28 | |
| M2OBackbone=Llama3.2-3B2026.06 | 27.7 | |
| SECBackbone=Llama3.2-3B2026.06 | 26.2 | |
| M2OBackbone=Qwen3-1.7B2026.06 | 25.8 | |
| BaseBackbone=Qwen3-1.7B2026.06 | 24.2 | |
| BaseBackbone=Llama3.2-3B2026.06 | 23.6 |