Трансформердің негізгі моделі

Fully non-autoregressive neural machine trans- lation (NAT) simultaneously predicts tokens with single forward of neural networks, which significantly reduces the inference latency at the expense of quality drop compared to …...

AbstractToken-level adaptive training approaches can alleviate the token imbalance problem and thus improve neural machine translation, through re-weighting the losses of different target tokens based on specific statistical metrics (e.g., …...