Language Modeling on WikiText-103 (Latency, Perplexity, and Efficiency Metrics)
14.6Latency (ms/token)HVT
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| HVT2025.07 | 14.6 | 2.3 | 16.7 | 1.3 | 72.4 | 62.7 | |
| DSBD2025.07 | 18.3 | 1.9 | 17 | 1.7 | 66.8 | 54.3 | |
| Speculative Decoding2025.07 | 21.8 | 1.6 | 17.9 | 2.1 | 63.2 | 48.5 | |
| Greedy Decoding2025.07 | 35.2 | 1 | 18.5 | 2.8 | — | — | |
| Multinomial Decoding2025.07 | 41.5 | 0.9 | 17.2 | 3.2 | — | — |