Mathematical Reasoning on GSM8K (Correct, Success, JTax)
89AccuracyBenchmark LLM (BenchAcc)
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Benchmark LLM (BenchAcc)2026.01 | 89 | — | — | |
| Gradient Shiftvariant=hybrid2026.01 | 88 | 99 | 0 | |
| Gradient Shiftvariant=mult2026.01 | 86 | 100 | 3 | |
| Weak-to-Strongalpha=2.0, k=402026.01 | 86 | 98 | 1 | |
| Weak-to-Strongalpha=1.5, k=402026.01 | 85 | 97 | 1 | |
| Helper LLM2026.01 | 85 | — | — | |
| Weak-to-Strongalpha=1.0, k=402026.01 | 81 | 98 | 7 | |
| Gradient Shiftvariant=add2026.01 | 58 | 70 | 6 |