# 基于Transformer的机器翻译实战 **Repository Path**: dot9527/Transformer ## Basic Information - **Project Name**: 基于Transformer的机器翻译实战 - **Description**: 基于Transformer的机器翻译实战 - **Primary Language**: Unknown - **License**: Not specified - **Default Branch**: master - **Homepage**: None - **GVP Project**: No ## Statistics - **Stars**: 0 - **Forks**: 10 - **Created**: 2025-12-04 - **Last Updated**: 2025-12-04 ## Categories & Tags **Categories**: Uncategorized **Tags**: None ## README ### 基于Transformer的机器翻译实战 - Transformer理论部分：https://blog.csdn.net/weixin_46649052/article/details/120050595 - Transformer代码：https://blog.csdn.net/weixin_46649052/article/details/120061870 这个项目与Transformer代码实现博客中的代码实现一样 环境配置为 - torch1.6(GPU) - python3.8 运行结果为： ```src_vocab 5493 tgt_vocab 2537 > start train mask = torch.nonzero(target.data == self.padding_idx) Epoch 0 Batch: 0 Loss: 7.869846 Tokens per Sec: 6.729871s Epoch 0 Batch: 50 Loss: 6.973424 Tokens per Sec: 9.369555s Epoch 0 Batch: 100 Loss: 6.151075 Tokens per Sec: 9.335995s >> Evaluate Epoch 0 Batch: 0 Loss: 5.811797 Tokens per Sec: 10.205480s <<<<< Evaluate loss: 5.894540 ****** Save model done... ****** Epoch 1 Batch: 0 Loss: 5.817729 Tokens per Sec: 9.177809s Epoch 1 Batch: 50 Loss: 5.375205 Tokens per Sec: 9.392804s Epoch 1 Batch: 100 Loss: 5.004299 Tokens per Sec: 9.312739s >> Evaluate Epoch 1 Batch: 0 Loss: 4.876106 Tokens per Sec: 10.117515s <<<<< Evaluate loss: 4.963982 ****** Save model done... ****** Epoch 2 Batch: 0 Loss: 4.972553 Tokens per Sec: 9.326845s Epoch 2 Batch: 50 Loss: 4.716051 Tokens per Sec: 9.343461s Epoch 2 Batch: 100 Loss: 4.421447 Tokens per Sec: 9.298382s >> Evaluate Epoch 2 Batch: 0 Loss: 4.322602 Tokens per Sec: 10.029434s <<<<< Evaluate loss: 4.404554 ****** Save model done... ****** Epoch 3 Batch: 0 Loss: 4.321961 Tokens per Sec: 9.205928s Epoch 3 Batch: 50 Loss: 4.124991 Tokens per Sec: 9.310113s Epoch 3 Batch: 100 Loss: 3.811377 Tokens per Sec: 9.258893s >> Evaluate Epoch 3 Batch: 0 Loss: 3.755665 Tokens per Sec: 9.987302s <<<<< Evaluate loss: 3.840265 ****** Save model done... ****** Epoch 4 Batch: 0 Loss: 3.806191 Tokens per Sec: 9.237126s Epoch 4 Batch: 50 Loss: 3.649720 Tokens per Sec: 9.304432s Epoch 4 Batch: 100 Loss: 3.349196 Tokens per Sec: 9.241163s >> Evaluate Epoch 4 Batch: 0 Loss: 3.337457 Tokens per Sec: 10.032519s <<<<< Evaluate loss: 3.417614 ****** Save model done... ****** Epoch 5 Batch: 0 Loss: 3.459634 Tokens per Sec: 9.176797s Epoch 5 Batch: 50 Loss: 3.273624 Tokens per Sec: 9.308859s Epoch 5 atch: 100 Loss: 2.998452 Tokens per Sec: 9.213062s >> Evaluate Epoch 5 Batch: 0 Loss: 2.960990 Tokens per Sec: 10.029608s <<<<< Evaluate loss: 3.052980 ****** Save model done... ****** Epoch 6 Batch: 0 Loss: 3.211699 Tokens per Sec: 9.185010s Epoch 6 Batch: 50 Loss: 2.893345 Tokens per Sec: 9.238220s Epoch 6 Batch: 100 Loss: 2.681960 Tokens per Sec: 9.188497s >> Evaluate Epoch 6 Batch: 0 Loss: 2.675523 Tokens per Sec: 9.989699s <<<<< Evaluate loss: 2.747432 ****** Save model done... ****** Epoch 7 Batch: 0 Loss: 2.950672 Tokens per Sec: 9.177439s Epoch 7 Batch: 50 Loss: 2.645654 Tokens per Sec: 9.243345s Epoch 7 Batch: 100 Loss: 2.393623 Tokens per Sec: 9.190061s >> Evaluate Epoch 7 Batch: 0 Loss: 2.380331 Tokens per Sec: 10.032775s <<<<< Evaluate loss: 2.472470 ****** Save model done... ****** Epoch 8 Batch: 0 Loss: 2.514863 Tokens per Sec: 9.237069s Epoch 8 Batch: 50 Loss: 2.363517 Tokens per Sec: 9.244510s Epoch 8 Batch: 100 Loss: 2.082449 Tokens per Sec: 9.155179s >> Evaluate Epoch 8 Batch: 0 Loss: 2.099774 Tokens per Sec: 10.031063s <<<<< Evaluate loss: 2.180049 ****** Save model done... ****** Epoch 9 Batch: 0 Loss: 2.320165 Tokens per Sec: 9.147264s Epoch 9 Batch: 50 Loss: 2.092441 Tokens per Sec: 9.262332s Epoch 9 Batch: 100 Loss: 1.818154 Tokens per Sec: 9.175017s >> Evaluate Epoch 9 Batch: 0 Loss: 1.842787 Tokens per Sec: 10.029404s <<<<< Evaluate loss: 1.941446 ****** Save model done... ****** Epoch 10 Batch: 0 Loss: 1.946543 Tokens per Sec: 8.972206s Epoch 10 Batch: 50 Loss: 1.917192 Tokens per Sec: 9.136343s Epoch 10 Batch: 100 Loss: 1.587525 Tokens per Sec: 9.199168s >> Evaluate Epoch 10 Batch: 0 Loss: 1.625920 Tokens per Sec: 9.988358s <<<<< Evaluate loss: 1.738675 ****** Save model done... ****** Epoch 11 Batch: 0 Loss: 1.732364 Tokens per Sec: 9.146457s Epoch 11 Batch: 50 Loss: 1.695014 Tokens per Sec: 9.187964s Epoch 11 Batch: 100 Loss: 1.454914 Tokens per Sec: 9.176912s >> Evaluate Epoch 11 Batch: 0 Loss: 1.514616 Tokens per Sec: 9.988358s <<<<< Evaluate loss: 1.647388 ****** Save model done... ****** Epoch 12 Batch: 0 Loss: 1.626071 Tokens per Sec: 9.059952s Epoch 12 Batch: 50 Loss: 1.541598 Tokens per Sec: 9.212269s Epoch 12 Batch: 100 Loss: 1.336604 Tokens per Sec: 9.202791s >> Evaluate Epoch 12 Batch: 0 Loss: 1.401758 Tokens per Sec: 9.862439s <<<<< Evaluate loss: 1.507253 ****** Save model done... ****** Epoch 13 Batch: 0 Loss: 1.420878 Tokens per Sec: 9.117863s Epoch 13 Batch: 50 Loss: 1.440541 Tokens per Sec: 9.189840s Epoch 13 Batch: 100 Loss: 1.132154 Tokens per Sec: 9.172929s >> Evaluate Epoch 13 Batch: 0 Loss: 1.207064 Tokens per Sec: 9.904041s <<<<< Evaluate loss: 1.320485 ****** Save model done... ****** Epoch 14 Batch: 0 Loss: 1.226669 Tokens per Sec: 9.118559s Epoch 14 Batch: 50 Loss: 1.261432 Tokens per Sec: 9.169248s Epoch 14 Batch: 100 Loss: 1.040873 Tokens per Sec: 9.119401s >> Evaluate Epoch 14 Batch: 0 Loss: 1.071591 Tokens per Sec: 9.821342s <<<<< Evaluate loss: 1.176238 ****** Save model done... ****** Epoch 15 Batch: 0 Loss: 1.086495 Tokens per Sec: 9.088573s Epoch 15 Batch: 50 Loss: 1.159208 Tokens per Sec: 9.132641s Epoch 15 Batch: 100 Loss: 0.943899 Tokens per Sec: 9.176910s >> Evaluate Epoch 15 Batch: 0 Loss: 0.962707 Tokens per Sec: 9.904061s <<<<< Evaluate loss: 1.063912 ****** Save model done... ****** Epoch 16 Batch: 0 Loss: 0.978801 Tokens per Sec: 9.002674s Epoch 16 Batch: 50 Loss: 1.079826 Tokens per Sec: 9.086403s Epoch 16 Batch: 100 Loss: 0.851728 Tokens per Sec: 9.041045s >> Evaluate Epoch 16 Batch: 0 Loss: 0.851689 Tokens per Sec: 9.241199s <<<<< Evaluate loss: 0.968978 ****** Save model done... ****** Epoch 17 Batch: 0 Loss: 0.919070 Tokens per Sec: 5.768479s Epoch 17 Batch: 50 Loss: 1.026209 Tokens per Sec: 5.864358s Epoch 17 Batch: 100 Loss: 0.791767 Tokens per Sec: 5.730444s >> Evaluate Epoch 17 Batch: 0 Loss: 0.779452 Tokens per Sec: 6.049646s <<<<< Evaluate loss: 0.875268 ****** Save model done... ****** Epoch 18 Batch: 0 Loss: 0.837940 Tokens per Sec: 5.839964s Epoch 18 Batch: 50 Loss: 0.900170 Tokens per Sec: 5.841982s Epoch 18 Batch: 100 Loss: 0.701193 Tokens per Sec: 5.800162s >> Evaluate Epoch 18 Batch: 0 Loss: 0.663479 Tokens per Sec: 6.065270s <<<<< Evaluate loss: 0.769979 ****** Save model done... ****** Epoch 19 Batch: 0 Loss: 0.751043 Tokens per Sec: 5.827926s Epoch 19 Batch: 50 Loss: 0.829178 Tokens per Sec: 5.816878s Epoch 19 Batch: 100 Loss: 0.635180 Tokens per Sec: 5.778595s >> Evaluate Epoch 19 Batch: 0 Loss: 0.614568 Tokens per Sec: 6.010801s <<<<< Evaluate loss: 0.695584 ****** Save model done... ****** <<<<<<< finished train, cost 418.6273 seconds > start predict BOS look around . EOS BOS 四 处 看 看 。 EOS translation: 继 续 看 。 BOS hurry up . EOS BOS 赶 快 ! EOS translation: 快 点 ！ BOS keep trying . EOS BOS 继 续 努 力 。 EOS translation: 继 续 考 试 。 BOS take it . EOS BOS 拿 走 吧 。 EOS translation: 拿 走 。 BOS birds fly . EOS BOS 鸟 类 飞 行 。 EOS translation: 鸟 类 穿 鸟 类 。 BOS hurry up . EOS BOS 快 点 ！ EOS translation: 快 点 ！ BOS look there . EOS BOS 看 那 里 。 EOS translation: 看 那 边 。```