The first language model explicitly declares each layer, while the second does the same with a loop.

basic_language_model.py

class LanguageModel(Module):
  """
  Takes three words as input and returns a probability for the next
  The 1st layer will use the first word's embedding
  The 2nd layer will use the 2nd word's embedding and the 1st word's output activations
  The 3rd layer will use the 3rd word's embedding plus the 2nd word's output activations
  """
  def __init__(self, vocab_sz, n_hidden):
    self.i_h = nn.Embedding(vocab_sz, n_hidden) # Converts the indices to a vector
    self.h_h = nn.Linear(n_hidden, n_hidden) # Creates the activations for the successive word
    self.h_o = nn.Linear(n_hidden, vocab_sz) # Predicts the fourth word
  
  def forward(self, x):
    h = F.relu(self.h_h(self.i_h(x[:, 0]))) # 1st word's embedding
    h += F.relu(self.i_h(x[:, 1])) # 2nd word's embedding
    h = F.relu(self.h_h(h)) # 1st word's output
    h += F.relu(self.i_h(x[:, 2])) # 3rd word's embedding
    h = F.relu(self.h_h(h)) # 2nd word's output
    return self.h_o(h)

class LanguageModelRecurrent(Module):
  """
  Instead of declaring the same layers three times, iterate over 3
  """
  def __init__(self, vocab_sz, n_hidden):
    self.i_h = nn.Embedding(vocab_sz, n_hidden)
    self.h_h = nn.Linear(n_hidden, n_hidden)
    self.h_o = nn.Linear(n_hidden, vocab_sz)
  
  def forward(self, x):
    h = 0
    for i in range(3):
      h += self.i_h(x[:, i])
      h = F.relu(self.h_h(h))
    return self.h_o(h)