Machine Translation on WMT 14 (BLEU, Rej., Speed)
27.5BLEUFaster Cascades
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Faster CascadesModel=T5 S→L, Decoding Setting=Greedy Decoding, Temperature=02026.06 | 27.5 | 35 | 1.85 | |
| VIA-SDModel=T5 S→XL, Decoding Setting=Greedy Decoding, Temperature=02026.06 | 23.1 | 27 | 3.35 | |
| VIA-SDModel=T5 S→XL, Decoding Setting=Non-Greedy Sampling, Temperature=12026.06 | 23.1 | 37 | 3.1 | |
| Faster CascadesModel=T5 S→XL, Decoding Setting=Greedy Decoding, Temperature=02026.06 | 23.05 | 31 | 2.7 | |
| Faster CascadesModel=T5 S→XL, Decoding Setting=Non-Greedy Sampling, Temperature=12026.06 | 23.05 | 39 | 2.55 | |
| Speculative DecodingModel=T5 S→XL, Decoding Setting=Non-Greedy Sampling, Temperature=12026.06 | 23 | 44 | 1.7 | |
| BiLDModel=T5 S→XL, Decoding Setting=Non-Greedy Sampling, Temperature=12026.06 | 23 | 42 | 1.8 | |
| Cascade SDModel=T5 S→XL, Decoding Setting=Non-Greedy Sampling, Temperature=12026.06 | 23 | 41 | 1.85 | |
| SWiFTModel=T5 S→XL, Decoding Setting=Non-Greedy Sampling, Temperature=12026.06 | 23 | 40 | 2 | |
| CLaSPModel=T5 S→XL, Decoding Setting=Non-Greedy Sampling, Temperature=12026.06 | 23 | 39 | 2.05 | |
| Speculative DecodingModel=T5 S→XL, Decoding Setting=Greedy Decoding, Temperature=02026.06 | 22.95 | 41 | 1.8 | |
| BiLDModel=T5 S→XL, Decoding Setting=Greedy Decoding, Temperature=02026.06 | 22.95 | 39 | 1.92 | |
| Cascade SDModel=T5 S→XL, Decoding Setting=Greedy Decoding, Temperature=02026.06 | 22.95 | 38 | 2 | |
| SWiFTModel=T5 S→XL, Decoding Setting=Greedy Decoding, Temperature=02026.06 | 22.95 | 34 | 2.2 | |
| CLaSPModel=T5 S→XL, Decoding Setting=Greedy Decoding, Temperature=02026.06 | 22.95 | 33 | 2.25 | |
| Faster CascadesModel=T5 S→L, Decoding Setting=Non-Greedy Sampling, Temperature=12026.06 | 22.5 | 36 | 1.78 | |
| VIA-SDModel=T5 S→L, Decoding Setting=Greedy Decoding, Temperature=02026.06 | 21.92 | 32 | 2.5 | |
| VIA-SDModel=T5 S→L, Decoding Setting=Non-Greedy Sampling, Temperature=12026.06 | 21.33 | 34 | 2.3 | |
| Speculative DecodingModel=T5 S→L, Decoding Setting=Non-Greedy Sampling, Temperature=12026.06 | 18.1 | 42 | 1.3 | |
| BiLDModel=T5 S→L, Decoding Setting=Non-Greedy Sampling, Temperature=12026.06 | 18.1 | 41 | 1.4 | |
| Cascade SDModel=T5 S→L, Decoding Setting=Non-Greedy Sampling, Temperature=12026.06 | 18.1 | 40 | 1.45 | |
| SWiFTModel=T5 S→L, Decoding Setting=Non-Greedy Sampling, Temperature=12026.06 | 18.1 | 38 | 1.55 | |
| CLaSPModel=T5 S→L, Decoding Setting=Non-Greedy Sampling, Temperature=12026.06 | 18.1 | 37 | 1.6 | |
| Speculative DecodingModel=T5 S→L, Decoding Setting=Greedy Decoding, Temperature=02026.06 | 18 | 39 | 1.4 | |
| BiLDModel=T5 S→L, Decoding Setting=Greedy Decoding, Temperature=02026.06 | 18 | 38 | 1.55 | |
| Cascade SDModel=T5 S→L, Decoding Setting=Greedy Decoding, Temperature=02026.06 | 18 | 38 | 1.55 | |
| SWiFTModel=T5 S→L, Decoding Setting=Greedy Decoding, Temperature=02026.06 | 18 | 37 | 1.7 | |
| CLaSPModel=T5 S→L, Decoding Setting=Greedy Decoding, Temperature=02026.06 | 18 | 36 | 1.75 |