The Seq2seq model has the characteristics that the encoder is used to analyze the input sequence, and the decoder is used to generate the output sequence, which is very suitable for the characteristics of machine translation tasks, making it achieve advanced performance in the field of machine translation. In addition, the attention mechanism has achieved remarkable results in many kinds of natural language processing tasks. The combination of seq2seq model and attention mechanism is the critical research direction in machine translation. To compare the effects of different attention mechanisms on the performance of machine translation models, this paper selects two classical attention mechanisms, including Synthetic Attention and Muti-head Attention, to test machine translation tasks. In addition, for another critical set of the sequence model, teacher-forcing, this paper also tests the influence of different teacher-forcing thresholds on the model performance. The experimental results show that Muti-head Attention has apparent advantages over Synthetic Attention. In addition, the reasonable choice of teacher-forcing also affects the stability of the model in the real environment.