Threshold Selection
Cross-source consensus on Threshold Selection from 1 sources and 5 claims.
1 sources · 5 claims
Dosage & preparation
Comparisons
Evidence quality
Highlighted claims
- The paper recommends theta 0.95 as a default because it balances quality and latency. — LEAP: Layer-wise Exit-Aware Pretraining for Efficient Transformer Inference
- A stricter training threshold than inference threshold is used to provide headroom under distribution shift. — LEAP: Layer-wise Exit-Aware Pretraining for Efficient Transformer Inference
- Quality-speed behavior was stable across a range of inference thresholds. — LEAP: Layer-wise Exit-Aware Pretraining for Efficient Transformer Inference
- A theta value of 0.99 is described as quality-critical but provides little layer reduction and poor wall-clock behavior in the decision table. — LEAP: Layer-wise Exit-Aware Pretraining for Efficient Transformer Inference
- A theta value of 0.90 is aggressive and trades higher layer reduction for higher nearest-neighbor failure. — LEAP: Layer-wise Exit-Aware Pretraining for Efficient Transformer Inference