Simultaneous Machine Translation on MuST-C en-es v1.0 (test)
35.14BLEUSimul-LLM
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Simul-LLMFine-tuning wait-k=7, Inference wait-k=7, Training samples=5M2023.12 | 35.14 | 6.73 | |
| NMT TransformerDecoding Scheme=non-simultaneous2023.12 | 32.64 | 23.1 | |
| Simul-LLMFine-tuning wait-k=7, Inference wait-k=3, Training samples=5M2023.12 | 30.31 | 3.34 | |
| NMT LLMDecoding Scheme=wait-32023.12 | 30.06 | 3.95 | |
| NMT LLM Single SBSwait-k value=3, beams=5, chunks=1, window size=62023.12 | 29.48 | 4.64 | |
| Simul-LLMFine-tuning wait-k=7, Inference wait-k=7, Training samples=2M2023.12 | 28.92 | 6.87 | |
| NMT LLM Chunk SBSwait-k value=7, beams=5, chunks=4, window size=202023.12 | 28.66 | 7.09 | |
| NMT LLM Single SBSwait-k value=3, beams=5, chunks=1, window size=102023.12 | 27.67 | 4.82 | |
| NMT LLM Chunk SBSwait-k value=5, beams=5, chunks=3, window size=152023.12 | 27 | 5.9 | |
| NMT LLM Chunk SBSwait-k value=3, beams=5, chunks=2, window size=102023.12 | 26.33 | 5.26 | |
| Simul-LLMFine-tuning wait-k=7, Inference wait-k=3, Training samples=2M2023.12 | 25.18 | 3.61 | |
| Monotonic TransformerWait-k=5, Decoding Scheme=SimulMT2023.12 | 24.9 | 2.58 | |
| Simul-LLMFine-tuning wait-k=3, Inference wait-k=3, Training samples=2M2023.12 | 23.68 | 3.64 |