STEM Reasoning on SuperGPQA (Pass@1)
32Pass@1 AccuracyTAC
Evaluation Results
| Method | Links | |
|---|---|---|
| TACBackbone=Qwen3-4B-Base, Seeds=12026.06 | 32 | |
| RandomBackbone=Qwen3-4B-Base, Seeds=12026.06 | 30.5 | |
| SECBackbone=Qwen3-4B-Base, Seeds=12026.06 | 29.9 | |
| TACBackbone=Qwen3-1.7B2026.06 | 22.1 | |
| RandomBackbone=Qwen3-1.7B2026.06 | 21.9 | |
| SECBackbone=Qwen3-1.7B2026.06 | 21.9 | |
| M2OBackbone=Qwen3-1.7B2026.06 | 21.7 | |
| M2OBackbone=Llama3.2-3B2026.06 | 21.5 | |
| TACBackbone=Llama3.2-3B2026.06 | 21.3 | |
| BaseBackbone=Llama3.2-3B2026.06 | 21 | |
| RandomBackbone=Llama3.2-3B2026.06 | 18.8 | |
| BaseBackbone=Qwen3-1.7B2026.06 | 18.5 | |
| SECBackbone=Llama3.2-3B2026.06 | 18 | |
| TACBackbone=Qwen3-0.6B-Base, Seeds=12026.06 | 16.9 | |
| RandomBackbone=Qwen3-0.6B-Base, Seeds=12026.06 | 15.6 | |
| SECBackbone=Qwen3-0.6B-Base, Seeds=12026.06 | 15 |