Reasoning Dataset
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
Reasoning Dataset Regime IV Collapse fallback-needed
1.09Speedup (Batch 1)
3
Reasoning Dataset Regime III: Process-degraded
2.21Speedup (batch-1)
3
Reasoning Dataset Regime II: Precision-sensitive
3.08Speedup (batch=1)
3
Reasoning Dataset Regime I: Low-bit-stable
3.4Speedup (batch=1)
3