Weird behavior using pytorch's nn.Embedding multiple times

Question

I want my model to take the input sequence as input 3 times. Once uni-gram, once bi-gram and once tri-gram. Each of the three will pass through it\'s own lstm and then it wi